projects / GitHub / mt8127 / android_kernel_alcatel_ttab.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
get rid of boilerplate switches in posix_acl.h
[GitHub/mt8127/android_kernel_alcatel_ttab.git] / fs / namei.c
CommitLineData
1da177e4
LT		 1/*
2 * linux/fs/namei.c
3 *
4 * Copyright (C) 1991, 1992 Linus Torvalds
5 */
6
7/*
8 * Some corrections by tytso.
9 */
10
11/* [Feb 1997 T. Schoebel-Theuer] Complete rewrite of the pathname
12 * lookup logic.
13 */
14/* [Feb-Apr 2000, AV] Rewrite to the new namespace architecture.
15 */
16
17#include <linux/init.h>
18#include <linux/module.h>
19#include <linux/slab.h>
20#include <linux/fs.h>
21#include <linux/namei.h>
1da177e4 22#include <linux/pagemap.h>
0eeca283 23#include <linux/fsnotify.h>
1da177e4
LT		 24#include <linux/personality.h>
25#include <linux/security.h>
6146f0d5 26#include <linux/ima.h>
1da177e4
LT		 27#include <linux/syscalls.h>
28#include <linux/mount.h>
29#include <linux/audit.h>
16f7e0fe 30#include <linux/capability.h>
834f2a4a 31#include <linux/file.h>
5590ff0d 32#include <linux/fcntl.h>
08ce5f16 33#include <linux/device_cgroup.h>
5ad4e53b 34#include <linux/fs_struct.h>
e77819e5 35#include <linux/posix_acl.h>
1da177e4
LT		 36#include <asm/uaccess.h>
37
e81e3f4d
EP		 38#include "internal.h"
39
1da177e4
LT		 40/* [Feb-1997 T. Schoebel-Theuer]
41 * Fundamental changes in the pathname lookup mechanisms (namei)
42 * were necessary because of omirr. The reason is that omirr needs
43 * to know the _real_ pathname, not the user-supplied one, in case
44 * of symlinks (and also when transname replacements occur).
45 *
46 * The new code replaces the old recursive symlink resolution with
47 * an iterative one (in case of non-nested symlink chains). It does
48 * this with calls to <fs>_follow_link().
49 * As a side effect, dir_namei(), _namei() and follow_link() are now
50 * replaced with a single function lookup_dentry() that can handle all
51 * the special cases of the former code.
52 *
53 * With the new dcache, the pathname is stored at each inode, at least as
54 * long as the refcount of the inode is positive. As a side effect, the
55 * size of the dcache depends on the inode cache and thus is dynamic.
56 *
57 * [29-Apr-1998 C. Scott Ananian] Updated above description of symlink
58 * resolution to correspond with current state of the code.
59 *
60 * Note that the symlink resolution is not *completely* iterative.
61 * There is still a significant amount of tail- and mid- recursion in
62 * the algorithm. Also, note that <fs>_readlink() is not used in
63 * lookup_dentry(): lookup_dentry() on the result of <fs>_readlink()
64 * may return different results than <fs>_follow_link(). Many virtual
65 * filesystems (including /proc) exhibit this behavior.
66 */
67
68/* [24-Feb-97 T. Schoebel-Theuer] Side effects caused by new implementation:
69 * New symlink semantics: when open() is called with flags O_CREAT | O_EXCL
70 * and the name already exists in form of a symlink, try to create the new
71 * name indicated by the symlink. The old code always complained that the
72 * name already exists, due to not following the symlink even if its target
73 * is nonexistent. The new semantics affects also mknod() and link() when
25985edc 74 * the name is a symlink pointing to a non-existent name.
1da177e4
LT		 75 *
76 * I don't know which semantics is the right one, since I have no access
77 * to standards. But I found by trial that HP-UX 9.0 has the full "new"
78 * semantics implemented, while SunOS 4.1.1 and Solaris (SunOS 5.4) have the
79 * "old" one. Personally, I think the new semantics is much more logical.
80 * Note that "ln old new" where "new" is a symlink pointing to a non-existing
81 * file does succeed in both HP-UX and SunOs, but not in Solaris
82 * and in the old Linux semantics.
83 */
84
85/* [16-Dec-97 Kevin Buhr] For security reasons, we change some symlink
86 * semantics. See the comments in "open_namei" and "do_link" below.
87 *
88 * [10-Sep-98 Alan Modra] Another symlink change.
89 */
90
91/* [Feb-Apr 2000 AV] Complete rewrite. Rules for symlinks:
92 * inside the path - always follow.
93 * in the last component in creation/removal/renaming - never follow.
94 * if LOOKUP_FOLLOW passed - follow.
95 * if the pathname has trailing slashes - follow.
96 * otherwise - don't follow.
97 * (applied in that order).
98 *
99 * [Jun 2000 AV] Inconsistent behaviour of open() in case if flags==O_CREAT
100 * restored for 2.4. This is the last surviving part of old 4.2BSD bug.
101 * During the 2.4 we need to fix the userland stuff depending on it -
102 * hopefully we will be able to get rid of that wart in 2.5. So far only
103 * XEmacs seems to be relying on it...
104 */
105/*
106 * [Sep 2001 AV] Single-semaphore locking scheme (kudos to David Holland)
a11f3a05 107 * implemented. Let's see if raised priority of ->s_vfs_rename_mutex gives
1da177e4
LT		 108 * any extra contention...
109 */
110
111/* In order to reduce some races, while at the same time doing additional
112 * checking and hopefully speeding things up, we copy filenames to the
113 * kernel data space before using them..
114 *
115 * POSIX.1 2.4: an empty pathname is invalid (ENOENT).
116 * PATH_MAX includes the nul terminator --RR.
117 */
858119e1 118static int do_getname(const char __user *filename, char *page)
1da177e4
LT		 119{
120 int retval;
121 unsigned long len = PATH_MAX;
122
123 if (!segment_eq(get_fs(), KERNEL_DS)) {
124 if ((unsigned long) filename >= TASK_SIZE)
125 return -EFAULT;
126 if (TASK_SIZE - (unsigned long) filename < PATH_MAX)
127 len = TASK_SIZE - (unsigned long) filename;
128 }
129
130 retval = strncpy_from_user(page, filename, len);
131 if (retval > 0) {
132 if (retval < len)
133 return 0;
134 return -ENAMETOOLONG;
135 } else if (!retval)
136 retval = -ENOENT;
137 return retval;
138}
139
f52e0c11 140static char *getname_flags(const char __user * filename, int flags)
1da177e4
LT		 141{
142 char *tmp, *result;
143
144 result = ERR_PTR(-ENOMEM);
145 tmp = __getname();
146 if (tmp) {
147 int retval = do_getname(filename, tmp);
148
149 result = tmp;
150 if (retval < 0) {
f52e0c11
AV		 151 if (retval != -ENOENT || !(flags & LOOKUP_EMPTY)) {
152 __putname(tmp);
153 result = ERR_PTR(retval);
154 }
1da177e4
LT		 155 }
156 }
157 audit_getname(result);
158 return result;
159}
160
f52e0c11
AV		 161char *getname(const char __user * filename)
162{
163 return getname_flags(filename, 0);
164}
165
1da177e4
LT		 166#ifdef CONFIG_AUDITSYSCALL
167void putname(const char *name)
168{
5ac3a9c2 169 if (unlikely(!audit_dummy_context()))
1da177e4
LT		 170 audit_putname(name);
171 else
172 __putname(name);
173}
174EXPORT_SYMBOL(putname);
175#endif
176
e77819e5
LT		 177static int check_acl(struct inode *inode, int mask)
178{
84635d68 179#ifdef CONFIG_FS_POSIX_ACL
e77819e5
LT		 180 struct posix_acl *acl;
181
182 /*
183 * Under RCU walk, we cannot even do a "get_cached_acl()",
184 * because that involves locking and getting a refcount on
185 * a cached ACL.
186 *
187 * So the only case we handle during RCU walking is the
188 * case of a cached "no ACL at all", which needs no locks
189 * or refcounts.
190 */
191 if (mask & MAY_NOT_BLOCK) {
192 if (negative_cached_acl(inode, ACL_TYPE_ACCESS))
193 return -EAGAIN;
194 return -ECHILD;
195 }
196
197 acl = get_cached_acl(inode, ACL_TYPE_ACCESS);
198
199 /*
4e34e719
CH		 200 * A filesystem can force a ACL callback by just never filling the
201 * ACL cache. But normally you'd fill the cache either at inode
202 * instantiation time, or on the first ->get_acl call.
e77819e5 203 *
4e34e719
CH		 204 * If the filesystem doesn't have a get_acl() function at all, we'll
205 * just create the negative cache entry.
e77819e5
LT		 206 */
207 if (acl == ACL_NOT_CACHED) {
4e34e719
CH		 208 if (inode->i_op->get_acl) {
209 acl = inode->i_op->get_acl(inode, ACL_TYPE_ACCESS);
210 if (IS_ERR(acl))
211 return PTR_ERR(acl);
212 } else {
213 set_cached_acl(inode, ACL_TYPE_ACCESS, NULL);
214 return -EAGAIN;
215 }
e77819e5
LT		 216 }
217
218 if (acl) {
219 int error = posix_acl_permission(inode, acl, mask);
220 posix_acl_release(acl);
221 return error;
222 }
84635d68 223#endif
e77819e5
LT		 224
225 return -EAGAIN;
226}
227
5909ccaa
LT		 228/*
229 * This does basic POSIX ACL permission checking
1da177e4 230 */
7e40145e 231static int acl_permission_check(struct inode *inode, int mask)
1da177e4 232{
26cf46be 233 unsigned int mode = inode->i_mode;
1da177e4 234
9c2c7039 235 mask &= MAY_READ | MAY_WRITE | MAY_EXEC | MAY_NOT_BLOCK;
e6305c43 236
e795b717
SH		 237 if (current_user_ns() != inode_userns(inode))
238 goto other_perms;
239
14067ff5 240 if (likely(current_fsuid() == inode->i_uid))
1da177e4
LT		 241 mode >>= 6;
242 else {
e77819e5 243 if (IS_POSIXACL(inode) && (mode & S_IRWXG)) {
7e40145e 244 int error = check_acl(inode, mask);
b74c79e9
NP		 245 if (error != -EAGAIN)
246 return error;
1da177e4
LT		 247 }
248
249 if (in_group_p(inode->i_gid))
250 mode >>= 3;
251 }
252
e795b717 253other_perms:
1da177e4
LT		 254 /*
255 * If the DACs are ok we don't need any capability check.
256 */
9c2c7039 257 if ((mask & ~mode & (MAY_READ | MAY_WRITE | MAY_EXEC)) == 0)
1da177e4 258 return 0;
5909ccaa
LT		 259 return -EACCES;
260}
261
262/**
b74c79e9 263 * generic_permission - check for access rights on a Posix-like filesystem
5909ccaa
LT		 264 * @inode: inode to check access rights for
265 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
5909ccaa
LT		 266 *
267 * Used to check for read/write/execute permissions on a file.
268 * We use "fsuid" for this, letting us set arbitrary permissions
269 * for filesystem access without changing the "normal" uids which
b74c79e9
NP		 270 * are used for other things.
271 *
272 * generic_permission is rcu-walk aware. It returns -ECHILD in case an rcu-walk
273 * request cannot be satisfied (eg. requires blocking or too much complexity).
274 * It would then be called again in ref-walk mode.
5909ccaa 275 */
2830ba7f 276int generic_permission(struct inode *inode, int mask)
5909ccaa
LT		 277{
278 int ret;
279
280 /*
281 * Do the basic POSIX ACL permission checks.
282 */
7e40145e 283 ret = acl_permission_check(inode, mask);
5909ccaa
LT		 284 if (ret != -EACCES)
285 return ret;
1da177e4 286
d594e7ec
AV		 287 if (S_ISDIR(inode->i_mode)) {
288 /* DACs are overridable for directories */
289 if (ns_capable(inode_userns(inode), CAP_DAC_OVERRIDE))
290 return 0;
291 if (!(mask & MAY_WRITE))
292 if (ns_capable(inode_userns(inode), CAP_DAC_READ_SEARCH))
293 return 0;
294 return -EACCES;
295 }
1da177e4
LT		 296 /*
297 * Read/write DACs are always overridable.
d594e7ec
AV		 298 * Executable DACs are overridable when there is
299 * at least one exec bit set.
1da177e4 300 */
d594e7ec 301 if (!(mask & MAY_EXEC) || (inode->i_mode & S_IXUGO))
e795b717 302 if (ns_capable(inode_userns(inode), CAP_DAC_OVERRIDE))
1da177e4
LT		 303 return 0;
304
305 /*
306 * Searching includes executable on directories, else just read.
307 */
7ea66001 308 mask &= MAY_READ | MAY_WRITE | MAY_EXEC;
d594e7ec 309 if (mask == MAY_READ)
e795b717 310 if (ns_capable(inode_userns(inode), CAP_DAC_READ_SEARCH))
1da177e4
LT		 311 return 0;
312
313 return -EACCES;
314}
315
cb23beb5
CH		 316/**
317 * inode_permission - check for access rights to a given inode
318 * @inode: inode to check permission on
319 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
320 *
321 * Used to check for read/write/execute permissions on an inode.
322 * We use "fsuid" for this, letting us set arbitrary permissions
323 * for filesystem access without changing the "normal" uids which
324 * are used for other things.
325 */
f419a2e3 326int inode_permission(struct inode *inode, int mask)
1da177e4 327{
e6305c43 328 int retval;
1da177e4
LT		 329
330 if (mask & MAY_WRITE) {
22590e41 331 umode_t mode = inode->i_mode;
1da177e4
LT		 332
333 /*
334 * Nobody gets write access to a read-only fs.
335 */
336 if (IS_RDONLY(inode) &&
337 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)))
338 return -EROFS;
339
340 /*
341 * Nobody gets write access to an immutable file.
342 */
343 if (IS_IMMUTABLE(inode))
344 return -EACCES;
345 }
346
acfa4380 347 if (inode->i_op->permission)
10556cb2 348 retval = inode->i_op->permission(inode, mask);
f696a365 349 else
2830ba7f 350 retval = generic_permission(inode, mask);
f696a365 351
1da177e4
LT		 352 if (retval)
353 return retval;
354
08ce5f16
SH		 355 retval = devcgroup_inode_permission(inode, mask);
356 if (retval)
357 return retval;
358
d09ca739 359 return security_inode_permission(inode, mask);
1da177e4
LT		 360}
361
5dd784d0
JB		 362/**
363 * path_get - get a reference to a path
364 * @path: path to get the reference to
365 *
366 * Given a path increment the reference count to the dentry and the vfsmount.
367 */
368void path_get(struct path *path)
369{
370 mntget(path->mnt);
371 dget(path->dentry);
372}
373EXPORT_SYMBOL(path_get);
374
1d957f9b
JB		 375/**
376 * path_put - put a reference to a path
377 * @path: path to put the reference to
378 *
379 * Given a path decrement the reference count to the dentry and the vfsmount.
380 */
381void path_put(struct path *path)
1da177e4 382{
1d957f9b
JB		 383 dput(path->dentry);
384 mntput(path->mnt);
1da177e4 385}
1d957f9b 386EXPORT_SYMBOL(path_put);
1da177e4 387
19660af7 388/*
31e6b01f 389 * Path walking has 2 modes, rcu-walk and ref-walk (see
19660af7
AV		 390 * Documentation/filesystems/path-lookup.txt). In situations when we can't
391 * continue in RCU mode, we attempt to drop out of rcu-walk mode and grab
392 * normal reference counts on dentries and vfsmounts to transition to rcu-walk
393 * mode. Refcounts are grabbed at the last known good point before rcu-walk
394 * got stuck, so ref-walk may continue from there. If this is not successful
395 * (eg. a seqcount has changed), then failure is returned and it's up to caller
396 * to restart the path walk from the beginning in ref-walk mode.
31e6b01f 397 */
31e6b01f
NP		 398
399/**
19660af7
AV		 400 * unlazy_walk - try to switch to ref-walk mode.
401 * @nd: nameidata pathwalk data
402 * @dentry: child of nd->path.dentry or NULL
39191628 403 * Returns: 0 on success, -ECHILD on failure
31e6b01f 404 *
19660af7
AV		 405 * unlazy_walk attempts to legitimize the current nd->path, nd->root and dentry
406 * for ref-walk mode. @dentry must be a path found by a do_lookup call on
407 * @nd or NULL. Must be called from rcu-walk context.
31e6b01f 408 */
19660af7 409static int unlazy_walk(struct nameidata *nd, struct dentry *dentry)
31e6b01f
NP		 410{
411 struct fs_struct *fs = current->fs;
412 struct dentry *parent = nd->path.dentry;
5b6ca027 413 int want_root = 0;
31e6b01f
NP		 414
415 BUG_ON(!(nd->flags & LOOKUP_RCU));
5b6ca027
AV		 416 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) {
417 want_root = 1;
31e6b01f
NP		 418 spin_lock(&fs->lock);
419 if (nd->root.mnt != fs->root.mnt ||
420 nd->root.dentry != fs->root.dentry)
421 goto err_root;
422 }
423 spin_lock(&parent->d_lock);
19660af7
AV		 424 if (!dentry) {
425 if (!__d_rcu_to_refcount(parent, nd->seq))
426 goto err_parent;
427 BUG_ON(nd->inode != parent->d_inode);
428 } else {
94c0d4ec
AV		 429 if (dentry->d_parent != parent)
430 goto err_parent;
19660af7
AV		 431 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
432 if (!__d_rcu_to_refcount(dentry, nd->seq))
433 goto err_child;
434 /*
435 * If the sequence check on the child dentry passed, then
436 * the child has not been removed from its parent. This
437 * means the parent dentry must be valid and able to take
438 * a reference at this point.
439 */
440 BUG_ON(!IS_ROOT(dentry) && dentry->d_parent != parent);
441 BUG_ON(!parent->d_count);
442 parent->d_count++;
443 spin_unlock(&dentry->d_lock);
444 }
31e6b01f 445 spin_unlock(&parent->d_lock);
5b6ca027 446 if (want_root) {
31e6b01f
NP		 447 path_get(&nd->root);
448 spin_unlock(&fs->lock);
449 }
450 mntget(nd->path.mnt);
451
452 rcu_read_unlock();
453 br_read_unlock(vfsmount_lock);
454 nd->flags &= ~LOOKUP_RCU;
455 return 0;
19660af7
AV		 456
457err_child:
31e6b01f 458 spin_unlock(&dentry->d_lock);
19660af7 459err_parent:
31e6b01f
NP		 460 spin_unlock(&parent->d_lock);
461err_root:
5b6ca027 462 if (want_root)
31e6b01f
NP		 463 spin_unlock(&fs->lock);
464 return -ECHILD;
465}
466
834f2a4a
TM		 467/**
468 * release_open_intent - free up open intent resources
469 * @nd: pointer to nameidata
470 */
471void release_open_intent(struct nameidata *nd)
472{
2dab5974
LT		 473 struct file *file = nd->intent.open.file;
474
475 if (file && !IS_ERR(file)) {
476 if (file->f_path.dentry == NULL)
477 put_filp(file);
478 else
479 fput(file);
480 }
834f2a4a
TM		 481}
482
f60aef7e 483static inline int d_revalidate(struct dentry *dentry, struct nameidata *nd)
34286d66 484{
f60aef7e 485 return dentry->d_op->d_revalidate(dentry, nd);
34286d66
NP		 486}
487
9f1fafee
AV		 488/**
489 * complete_walk - successful completion of path walk
490 * @nd: pointer nameidata
39159de2 491 *
9f1fafee
AV		 492 * If we had been in RCU mode, drop out of it and legitimize nd->path.
493 * Revalidate the final result, unless we'd already done that during
494 * the path walk or the filesystem doesn't ask for it. Return 0 on
495 * success, -error on failure. In case of failure caller does not
496 * need to drop nd->path.
39159de2 497 */
9f1fafee 498static int complete_walk(struct nameidata *nd)
39159de2 499{
16c2cd71 500 struct dentry *dentry = nd->path.dentry;
39159de2 501 int status;
39159de2 502
9f1fafee
AV		 503 if (nd->flags & LOOKUP_RCU) {
504 nd->flags &= ~LOOKUP_RCU;
505 if (!(nd->flags & LOOKUP_ROOT))
506 nd->root.mnt = NULL;
507 spin_lock(&dentry->d_lock);
508 if (unlikely(!__d_rcu_to_refcount(dentry, nd->seq))) {
509 spin_unlock(&dentry->d_lock);
510 rcu_read_unlock();
511 br_read_unlock(vfsmount_lock);
512 return -ECHILD;
513 }
514 BUG_ON(nd->inode != dentry->d_inode);
515 spin_unlock(&dentry->d_lock);
516 mntget(nd->path.mnt);
517 rcu_read_unlock();
518 br_read_unlock(vfsmount_lock);
519 }
520
16c2cd71
AV		 521 if (likely(!(nd->flags & LOOKUP_JUMPED)))
522 return 0;
523
524 if (likely(!(dentry->d_flags & DCACHE_OP_REVALIDATE)))
39159de2
JL		 525 return 0;
526
16c2cd71
AV		 527 if (likely(!(dentry->d_sb->s_type->fs_flags & FS_REVAL_DOT)))
528 return 0;
529
530 /* Note: we do not d_invalidate() */
34286d66 531 status = d_revalidate(dentry, nd);
39159de2
JL		 532 if (status > 0)
533 return 0;
534
16c2cd71 535 if (!status)
39159de2 536 status = -ESTALE;
16c2cd71 537
9f1fafee 538 path_put(&nd->path);
39159de2
JL		 539 return status;
540}
541
2a737871
AV		 542static __always_inline void set_root(struct nameidata *nd)
543{
f7ad3c6b
MS		 544 if (!nd->root.mnt)
545 get_fs_root(current->fs, &nd->root);
2a737871
AV		 546}
547
6de88d72
AV		 548static int link_path_walk(const char *, struct nameidata *);
549
31e6b01f
NP		 550static __always_inline void set_root_rcu(struct nameidata *nd)
551{
552 if (!nd->root.mnt) {
553 struct fs_struct *fs = current->fs;
c28cc364
NP		 554 unsigned seq;
555
556 do {
557 seq = read_seqcount_begin(&fs->seq);
558 nd->root = fs->root;
c1530019 559 nd->seq = __read_seqcount_begin(&nd->root.dentry->d_seq);
c28cc364 560 } while (read_seqcount_retry(&fs->seq, seq));
31e6b01f
NP		 561 }
562}
563
f1662356 564static __always_inline int __vfs_follow_link(struct nameidata *nd, const char *link)
1da177e4 565{
31e6b01f
NP		 566 int ret;
567
1da177e4
LT		 568 if (IS_ERR(link))
569 goto fail;
570
571 if (*link == '/') {
2a737871 572 set_root(nd);
1d957f9b 573 path_put(&nd->path);
2a737871
AV		 574 nd->path = nd->root;
575 path_get(&nd->root);
16c2cd71 576 nd->flags |= LOOKUP_JUMPED;
1da177e4 577 }
31e6b01f 578 nd->inode = nd->path.dentry->d_inode;
b4091d5f 579
31e6b01f
NP		 580 ret = link_path_walk(link, nd);
581 return ret;
1da177e4 582fail:
1d957f9b 583 path_put(&nd->path);
1da177e4
LT		 584 return PTR_ERR(link);
585}
586
1d957f9b 587static void path_put_conditional(struct path *path, struct nameidata *nd)
051d3812
IK		 588{
589 dput(path->dentry);
4ac91378 590 if (path->mnt != nd->path.mnt)
051d3812
IK		 591 mntput(path->mnt);
592}
593
7b9337aa
NP		 594static inline void path_to_nameidata(const struct path *path,
595 struct nameidata *nd)
051d3812 596{
31e6b01f
NP		 597 if (!(nd->flags & LOOKUP_RCU)) {
598 dput(nd->path.dentry);
599 if (nd->path.mnt != path->mnt)
600 mntput(nd->path.mnt);
9a229683 601 }
31e6b01f 602 nd->path.mnt = path->mnt;
4ac91378 603 nd->path.dentry = path->dentry;
051d3812
IK		 604}
605
574197e0
AV		 606static inline void put_link(struct nameidata *nd, struct path *link, void *cookie)
607{
608 struct inode *inode = link->dentry->d_inode;
609 if (!IS_ERR(cookie) && inode->i_op->put_link)
610 inode->i_op->put_link(link->dentry, nd, cookie);
611 path_put(link);
612}
613
def4af30 614static __always_inline int
574197e0 615follow_link(struct path *link, struct nameidata *nd, void **p)
1da177e4
LT		 616{
617 int error;
7b9337aa 618 struct dentry *dentry = link->dentry;
1da177e4 619
844a3917
AV		 620 BUG_ON(nd->flags & LOOKUP_RCU);
621
0e794589
AV		 622 if (link->mnt == nd->path.mnt)
623 mntget(link->mnt);
624
574197e0
AV		 625 if (unlikely(current->total_link_count >= 40)) {
626 *p = ERR_PTR(-ELOOP); /* no ->put_link(), please */
574197e0
AV		 627 path_put(&nd->path);
628 return -ELOOP;
629 }
630 cond_resched();
631 current->total_link_count++;
632
7b9337aa 633 touch_atime(link->mnt, dentry);
1da177e4 634 nd_set_link(nd, NULL);
cd4e91d3 635
36f3b4f6
AV		 636 error = security_inode_follow_link(link->dentry, nd);
637 if (error) {
638 *p = ERR_PTR(error); /* no ->put_link(), please */
639 path_put(&nd->path);
640 return error;
641 }
642
86acdca1 643 nd->last_type = LAST_BIND;
def4af30
AV		 644 *p = dentry->d_inode->i_op->follow_link(dentry, nd);
645 error = PTR_ERR(*p);
646 if (!IS_ERR(*p)) {
1da177e4 647 char *s = nd_get_link(nd);
cc314eef 648 error = 0;
1da177e4
LT		 649 if (s)
650 error = __vfs_follow_link(nd, s);
bcda7652 651 else if (nd->last_type == LAST_BIND) {
16c2cd71 652 nd->flags |= LOOKUP_JUMPED;
b21041d0
AV		 653 nd->inode = nd->path.dentry->d_inode;
654 if (nd->inode->i_op->follow_link) {
bcda7652
AV		 655 /* stepped on a _really_ weird one */
656 path_put(&nd->path);
657 error = -ELOOP;
658 }
659 }
1da177e4 660 }
1da177e4
LT		 661 return error;
662}
663
31e6b01f
NP		 664static int follow_up_rcu(struct path *path)
665{
666 struct vfsmount *parent;
667 struct dentry *mountpoint;
668
669 parent = path->mnt->mnt_parent;
670 if (parent == path->mnt)
671 return 0;
672 mountpoint = path->mnt->mnt_mountpoint;
673 path->dentry = mountpoint;
674 path->mnt = parent;
675 return 1;
676}
677
bab77ebf 678int follow_up(struct path *path)
1da177e4
LT		 679{
680 struct vfsmount *parent;
681 struct dentry *mountpoint;
99b7db7b
NP		 682
683 br_read_lock(vfsmount_lock);
bab77ebf
AV		 684 parent = path->mnt->mnt_parent;
685 if (parent == path->mnt) {
99b7db7b 686 br_read_unlock(vfsmount_lock);
1da177e4
LT		 687 return 0;
688 }
689 mntget(parent);
bab77ebf 690 mountpoint = dget(path->mnt->mnt_mountpoint);
99b7db7b 691 br_read_unlock(vfsmount_lock);
bab77ebf
AV		 692 dput(path->dentry);
693 path->dentry = mountpoint;
694 mntput(path->mnt);
695 path->mnt = parent;
1da177e4
LT		 696 return 1;
697}
698
b5c84bf6 699/*
9875cf80
DH		 700 * Perform an automount
701 * - return -EISDIR to tell follow_managed() to stop and return the path we
702 * were called with.
1da177e4 703 */
9875cf80
DH		 704static int follow_automount(struct path *path, unsigned flags,
705 bool *need_mntput)
31e6b01f 706{
9875cf80 707 struct vfsmount *mnt;
ea5b778a 708 int err;
9875cf80
DH		 709
710 if (!path->dentry->d_op || !path->dentry->d_op->d_automount)
711 return -EREMOTE;
712
6f45b656
DH		 713 /* We don't want to mount if someone supplied AT_NO_AUTOMOUNT
714 * and this is the terminal part of the path.
715 */
49084c3b 716 if ((flags & LOOKUP_NO_AUTOMOUNT) && !(flags & LOOKUP_PARENT))
6f45b656
DH		 717 return -EISDIR; /* we actually want to stop here */
718
5a30d8a2 719 /*
9875cf80
DH		 720 * We don't want to mount if someone's just doing a stat and they've
721 * set AT_SYMLINK_NOFOLLOW - unless they're stat'ing a directory and
722 * appended a '/' to the name.
723 */
5a30d8a2
DH		 724 if (!(flags & LOOKUP_FOLLOW)) {
725 /* We do, however, want to mount if someone wants to open or
726 * create a file of any type under the mountpoint, wants to
727 * traverse through the mountpoint or wants to open the mounted
728 * directory.
729 * Also, autofs may mark negative dentries as being automount
730 * points. These will need the attentions of the daemon to
731 * instantiate them before they can be used.
732 */
733 if (!(flags & (LOOKUP_PARENT | LOOKUP_DIRECTORY |
734 LOOKUP_OPEN | LOOKUP_CREATE)) &&
735 path->dentry->d_inode)
736 return -EISDIR;
737 }
9875cf80
DH		 738 current->total_link_count++;
739 if (current->total_link_count >= 40)
740 return -ELOOP;
741
742 mnt = path->dentry->d_op->d_automount(path);
743 if (IS_ERR(mnt)) {
744 /*
745 * The filesystem is allowed to return -EISDIR here to indicate
746 * it doesn't want to automount. For instance, autofs would do
747 * this so that its userspace daemon can mount on this dentry.
748 *
749 * However, we can only permit this if it's a terminal point in
750 * the path being looked up; if it wasn't then the remainder of
751 * the path is inaccessible and we should say so.
752 */
49084c3b 753 if (PTR_ERR(mnt) == -EISDIR && (flags & LOOKUP_PARENT))
9875cf80
DH		 754 return -EREMOTE;
755 return PTR_ERR(mnt);
31e6b01f 756 }
ea5b778a 757
9875cf80
DH		 758 if (!mnt) /* mount collision */
759 return 0;
31e6b01f 760
8aef1884
AV		 761 if (!*need_mntput) {
762 /* lock_mount() may release path->mnt on error */
763 mntget(path->mnt);
764 *need_mntput = true;
765 }
19a167af 766 err = finish_automount(mnt, path);
9875cf80 767
ea5b778a
DH		 768 switch (err) {
769 case -EBUSY:
770 /* Someone else made a mount here whilst we were busy */
19a167af 771 return 0;
ea5b778a 772 case 0:
8aef1884 773 path_put(path);
ea5b778a
DH		 774 path->mnt = mnt;
775 path->dentry = dget(mnt->mnt_root);
ea5b778a 776 return 0;
19a167af
AV		 777 default:
778 return err;
ea5b778a 779 }
19a167af 780
463ffb2e
AV		 781}
782
9875cf80
DH		 783/*
784 * Handle a dentry that is managed in some way.
cc53ce53 785 * - Flagged for transit management (autofs)
9875cf80
DH		 786 * - Flagged as mountpoint
787 * - Flagged as automount point
788 *
789 * This may only be called in refwalk mode.
790 *
791 * Serialization is taken care of in namespace.c
792 */
793static int follow_managed(struct path *path, unsigned flags)
1da177e4 794{
8aef1884 795 struct vfsmount *mnt = path->mnt; /* held by caller, must be left alone */
9875cf80
DH		 796 unsigned managed;
797 bool need_mntput = false;
8aef1884 798 int ret = 0;
9875cf80
DH		 799
800 /* Given that we're not holding a lock here, we retain the value in a
801 * local variable for each dentry as we look at it so that we don't see
802 * the components of that value change under us */
803 while (managed = ACCESS_ONCE(path->dentry->d_flags),
804 managed &= DCACHE_MANAGED_DENTRY,
805 unlikely(managed != 0)) {
cc53ce53
DH		 806 /* Allow the filesystem to manage the transit without i_mutex
807 * being held. */
808 if (managed & DCACHE_MANAGE_TRANSIT) {
809 BUG_ON(!path->dentry->d_op);
810 BUG_ON(!path->dentry->d_op->d_manage);
1aed3e42 811 ret = path->dentry->d_op->d_manage(path->dentry, false);
cc53ce53 812 if (ret < 0)
8aef1884 813 break;
cc53ce53
DH		 814 }
815
9875cf80
DH		 816 /* Transit to a mounted filesystem. */
817 if (managed & DCACHE_MOUNTED) {
818 struct vfsmount *mounted = lookup_mnt(path);
819 if (mounted) {
820 dput(path->dentry);
821 if (need_mntput)
822 mntput(path->mnt);
823 path->mnt = mounted;
824 path->dentry = dget(mounted->mnt_root);
825 need_mntput = true;
826 continue;
827 }
828
829 /* Something is mounted on this dentry in another
830 * namespace and/or whatever was mounted there in this
831 * namespace got unmounted before we managed to get the
832 * vfsmount_lock */
833 }
834
835 /* Handle an automount point */
836 if (managed & DCACHE_NEED_AUTOMOUNT) {
837 ret = follow_automount(path, flags, &need_mntput);
838 if (ret < 0)
8aef1884 839 break;
9875cf80
DH		 840 continue;
841 }
842
843 /* We didn't change the current path point */
844 break;
1da177e4 845 }
8aef1884
AV		 846
847 if (need_mntput && path->mnt == mnt)
848 mntput(path->mnt);
849 if (ret == -EISDIR)
850 ret = 0;
851 return ret;
1da177e4
LT		 852}
853
cc53ce53 854int follow_down_one(struct path *path)
1da177e4
LT		 855{
856 struct vfsmount *mounted;
857
1c755af4 858 mounted = lookup_mnt(path);
1da177e4 859 if (mounted) {
9393bd07
AV		 860 dput(path->dentry);
861 mntput(path->mnt);
862 path->mnt = mounted;
863 path->dentry = dget(mounted->mnt_root);
1da177e4
LT		 864 return 1;
865 }
866 return 0;
867}
868
62a7375e
IK		 869static inline bool managed_dentry_might_block(struct dentry *dentry)
870{
871 return (dentry->d_flags & DCACHE_MANAGE_TRANSIT &&
872 dentry->d_op->d_manage(dentry, true) < 0);
873}
874
9875cf80 875/*
287548e4
AV		 876 * Try to skip to top of mountpoint pile in rcuwalk mode. Fail if
877 * we meet a managed dentry that would need blocking.
9875cf80
DH		 878 */
879static bool __follow_mount_rcu(struct nameidata *nd, struct path *path,
287548e4 880 struct inode **inode)
9875cf80 881{
62a7375e 882 for (;;) {
9875cf80 883 struct vfsmount *mounted;
62a7375e
IK		 884 /*
885 * Don't forget we might have a non-mountpoint managed dentry
886 * that wants to block transit.
887 */
287548e4 888 if (unlikely(managed_dentry_might_block(path->dentry)))
ab90911f 889 return false;
62a7375e
IK		 890
891 if (!d_mountpoint(path->dentry))
892 break;
893
9875cf80
DH		 894 mounted = __lookup_mnt(path->mnt, path->dentry, 1);
895 if (!mounted)
896 break;
897 path->mnt = mounted;
898 path->dentry = mounted->mnt_root;
899 nd->seq = read_seqcount_begin(&path->dentry->d_seq);
59430262
LT		 900 /*
901 * Update the inode too. We don't need to re-check the
902 * dentry sequence number here after this d_inode read,
903 * because a mount-point is always pinned.
904 */
905 *inode = path->dentry->d_inode;
9875cf80 906 }
9875cf80
DH		 907 return true;
908}
909
dea39376 910static void follow_mount_rcu(struct nameidata *nd)
287548e4 911{
dea39376 912 while (d_mountpoint(nd->path.dentry)) {
287548e4 913 struct vfsmount *mounted;
dea39376 914 mounted = __lookup_mnt(nd->path.mnt, nd->path.dentry, 1);
287548e4
AV		 915 if (!mounted)
916 break;
dea39376
AV		 917 nd->path.mnt = mounted;
918 nd->path.dentry = mounted->mnt_root;
919 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
287548e4
AV		 920 }
921}
922
31e6b01f
NP		 923static int follow_dotdot_rcu(struct nameidata *nd)
924{
31e6b01f
NP		 925 set_root_rcu(nd);
926
9875cf80 927 while (1) {
31e6b01f
NP		 928 if (nd->path.dentry == nd->root.dentry &&
929 nd->path.mnt == nd->root.mnt) {
930 break;
931 }
932 if (nd->path.dentry != nd->path.mnt->mnt_root) {
933 struct dentry *old = nd->path.dentry;
934 struct dentry *parent = old->d_parent;
935 unsigned seq;
936
937 seq = read_seqcount_begin(&parent->d_seq);
938 if (read_seqcount_retry(&old->d_seq, nd->seq))
ef7562d5 939 goto failed;
31e6b01f
NP		 940 nd->path.dentry = parent;
941 nd->seq = seq;
942 break;
943 }
944 if (!follow_up_rcu(&nd->path))
945 break;
946 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
31e6b01f 947 }
dea39376
AV		 948 follow_mount_rcu(nd);
949 nd->inode = nd->path.dentry->d_inode;
31e6b01f 950 return 0;
ef7562d5
AV		 951
952failed:
953 nd->flags &= ~LOOKUP_RCU;
5b6ca027
AV		 954 if (!(nd->flags & LOOKUP_ROOT))
955 nd->root.mnt = NULL;
ef7562d5
AV		 956 rcu_read_unlock();
957 br_read_unlock(vfsmount_lock);
958 return -ECHILD;
31e6b01f
NP		 959}
960
cc53ce53
DH		 961/*
962 * Follow down to the covering mount currently visible to userspace. At each
963 * point, the filesystem owning that dentry may be queried as to whether the
964 * caller is permitted to proceed or not.
cc53ce53 965 */
7cc90cc3 966int follow_down(struct path *path)
cc53ce53
DH		 967{
968 unsigned managed;
969 int ret;
970
971 while (managed = ACCESS_ONCE(path->dentry->d_flags),
972 unlikely(managed & DCACHE_MANAGED_DENTRY)) {
973 /* Allow the filesystem to manage the transit without i_mutex
974 * being held.
975 *
976 * We indicate to the filesystem if someone is trying to mount
977 * something here. This gives autofs the chance to deny anyone
978 * other than its daemon the right to mount on its
979 * superstructure.
980 *
981 * The filesystem may sleep at this point.
982 */
983 if (managed & DCACHE_MANAGE_TRANSIT) {
984 BUG_ON(!path->dentry->d_op);
985 BUG_ON(!path->dentry->d_op->d_manage);
ab90911f 986 ret = path->dentry->d_op->d_manage(
1aed3e42 987 path->dentry, false);
cc53ce53
DH		 988 if (ret < 0)
989 return ret == -EISDIR ? 0 : ret;
990 }
991
992 /* Transit to a mounted filesystem. */
993 if (managed & DCACHE_MOUNTED) {
994 struct vfsmount *mounted = lookup_mnt(path);
995 if (!mounted)
996 break;
997 dput(path->dentry);
998 mntput(path->mnt);
999 path->mnt = mounted;
1000 path->dentry = dget(mounted->mnt_root);
1001 continue;
1002 }
1003
1004 /* Don't handle automount points here */
1005 break;
1006 }
1007 return 0;
1008}
1009
9875cf80
DH		 1010/*
1011 * Skip to top of mountpoint pile in refwalk mode for follow_dotdot()
1012 */
1013static void follow_mount(struct path *path)
1014{
1015 while (d_mountpoint(path->dentry)) {
1016 struct vfsmount *mounted = lookup_mnt(path);
1017 if (!mounted)
1018 break;
1019 dput(path->dentry);
1020 mntput(path->mnt);
1021 path->mnt = mounted;
1022 path->dentry = dget(mounted->mnt_root);
1023 }
1024}
1025
31e6b01f 1026static void follow_dotdot(struct nameidata *nd)
1da177e4 1027{
2a737871 1028 set_root(nd);
e518ddb7 1029
1da177e4 1030 while(1) {
4ac91378 1031 struct dentry *old = nd->path.dentry;
1da177e4 1032
2a737871
AV		 1033 if (nd->path.dentry == nd->root.dentry &&
1034 nd->path.mnt == nd->root.mnt) {
1da177e4
LT		 1035 break;
1036 }
4ac91378 1037 if (nd->path.dentry != nd->path.mnt->mnt_root) {
3088dd70
AV		 1038 /* rare case of legitimate dget_parent()... */
1039 nd->path.dentry = dget_parent(nd->path.dentry);
1da177e4
LT		 1040 dput(old);
1041 break;
1042 }
3088dd70 1043 if (!follow_up(&nd->path))
1da177e4 1044 break;
1da177e4 1045 }
79ed0226 1046 follow_mount(&nd->path);
31e6b01f 1047 nd->inode = nd->path.dentry->d_inode;
1da177e4
LT		 1048}
1049
baa03890
NP		 1050/*
1051 * Allocate a dentry with name and parent, and perform a parent
1052 * directory ->lookup on it. Returns the new dentry, or ERR_PTR
1053 * on error. parent->d_inode->i_mutex must be held. d_lookup must
1054 * have verified that no child exists while under i_mutex.
1055 */
1056static struct dentry *d_alloc_and_lookup(struct dentry *parent,
1057 struct qstr *name, struct nameidata *nd)
1058{
1059 struct inode *inode = parent->d_inode;
1060 struct dentry *dentry;
1061 struct dentry *old;
1062
1063 /* Don't create child dentry for a dead directory. */
1064 if (unlikely(IS_DEADDIR(inode)))
1065 return ERR_PTR(-ENOENT);
1066
1067 dentry = d_alloc(parent, name);
1068 if (unlikely(!dentry))
1069 return ERR_PTR(-ENOMEM);
1070
1071 old = inode->i_op->lookup(inode, dentry, nd);
1072 if (unlikely(old)) {
1073 dput(dentry);
1074 dentry = old;
1075 }
1076 return dentry;
1077}
1078
44396f4b
JB		 1079/*
1080 * We already have a dentry, but require a lookup to be performed on the parent
1081 * directory to fill in d_inode. Returns the new dentry, or ERR_PTR on error.
1082 * parent->d_inode->i_mutex must be held. d_lookup must have verified that no
1083 * child exists while under i_mutex.
1084 */
1085static struct dentry *d_inode_lookup(struct dentry *parent, struct dentry *dentry,
1086 struct nameidata *nd)
1087{
1088 struct inode *inode = parent->d_inode;
1089 struct dentry *old;
1090
1091 /* Don't create child dentry for a dead directory. */
1092 if (unlikely(IS_DEADDIR(inode)))
1093 return ERR_PTR(-ENOENT);
1094
1095 old = inode->i_op->lookup(inode, dentry, nd);
1096 if (unlikely(old)) {
1097 dput(dentry);
1098 dentry = old;
1099 }
1100 return dentry;
1101}
1102
1da177e4
LT		 1103/*
1104 * It's more convoluted than I'd like it to be, but... it's still fairly
1105 * small and for now I'd prefer to have fast path as straight as possible.
1106 * It _is_ time-critical.
1107 */
1108static int do_lookup(struct nameidata *nd, struct qstr *name,
31e6b01f 1109 struct path *path, struct inode **inode)
1da177e4 1110{
4ac91378 1111 struct vfsmount *mnt = nd->path.mnt;
31e6b01f 1112 struct dentry *dentry, *parent = nd->path.dentry;
5a18fff2
AV		 1113 int need_reval = 1;
1114 int status = 1;
9875cf80
DH		 1115 int err;
1116
b04f784e
NP		 1117 /*
1118 * Rename seqlock is not required here because in the off chance
1119 * of a false negative due to a concurrent rename, we're going to
1120 * do the non-racy lookup, below.
1121 */
31e6b01f
NP		 1122 if (nd->flags & LOOKUP_RCU) {
1123 unsigned seq;
31e6b01f
NP		 1124 *inode = nd->inode;
1125 dentry = __d_lookup_rcu(parent, name, &seq, inode);
5a18fff2
AV		 1126 if (!dentry)
1127 goto unlazy;
1128
31e6b01f
NP		 1129 /* Memory barrier in read_seqcount_begin of child is enough */
1130 if (__read_seqcount_retry(&parent->d_seq, nd->seq))
1131 return -ECHILD;
31e6b01f 1132 nd->seq = seq;
5a18fff2 1133
24643087 1134 if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE)) {
5a18fff2
AV		 1135 status = d_revalidate(dentry, nd);
1136 if (unlikely(status <= 0)) {
1137 if (status != -ECHILD)
1138 need_reval = 0;
1139 goto unlazy;
1140 }
24643087 1141 }
44396f4b
JB		 1142 if (unlikely(d_need_lookup(dentry)))
1143 goto unlazy;
31e6b01f
NP		 1144 path->mnt = mnt;
1145 path->dentry = dentry;
d6e9bd25
AV		 1146 if (unlikely(!__follow_mount_rcu(nd, path, inode)))
1147 goto unlazy;
1148 if (unlikely(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT))
1149 goto unlazy;
1150 return 0;
5a18fff2 1151unlazy:
19660af7
AV		 1152 if (unlazy_walk(nd, dentry))
1153 return -ECHILD;
5a18fff2
AV		 1154 } else {
1155 dentry = __d_lookup(parent, name);
9875cf80 1156 }
5a18fff2 1157
44396f4b
JB		 1158 if (dentry && unlikely(d_need_lookup(dentry))) {
1159 dput(dentry);
1160 dentry = NULL;
1161 }
5a18fff2
AV		 1162retry:
1163 if (unlikely(!dentry)) {
1164 struct inode *dir = parent->d_inode;
1165 BUG_ON(nd->inode != dir);
1166
1167 mutex_lock(&dir->i_mutex);
1168 dentry = d_lookup(parent, name);
1169 if (likely(!dentry)) {
1170 dentry = d_alloc_and_lookup(parent, name, nd);
1171 if (IS_ERR(dentry)) {
1172 mutex_unlock(&dir->i_mutex);
1173 return PTR_ERR(dentry);
1174 }
1175 /* known good */
1176 need_reval = 0;
1177 status = 1;
44396f4b
JB		 1178 } else if (unlikely(d_need_lookup(dentry))) {
1179 dentry = d_inode_lookup(parent, dentry, nd);
1180 if (IS_ERR(dentry)) {
1181 mutex_unlock(&dir->i_mutex);
1182 return PTR_ERR(dentry);
1183 }
1184 /* known good */
1185 need_reval = 0;
1186 status = 1;
5a18fff2
AV		 1187 }
1188 mutex_unlock(&dir->i_mutex);
1189 }
1190 if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE) && need_reval)
1191 status = d_revalidate(dentry, nd);
1192 if (unlikely(status <= 0)) {
1193 if (status < 0) {
1194 dput(dentry);
1195 return status;
1196 }
1197 if (!d_invalidate(dentry)) {
1198 dput(dentry);
1199 dentry = NULL;
1200 need_reval = 1;
1201 goto retry;
1202 }
24643087 1203 }
5a18fff2 1204
9875cf80
DH		 1205 path->mnt = mnt;
1206 path->dentry = dentry;
1207 err = follow_managed(path, nd->flags);
89312214
IK		 1208 if (unlikely(err < 0)) {
1209 path_put_conditional(path, nd);
9875cf80 1210 return err;
89312214 1211 }
9875cf80 1212 *inode = path->dentry->d_inode;
1da177e4 1213 return 0;
1da177e4
LT		 1214}
1215
52094c8a
AV		 1216static inline int may_lookup(struct nameidata *nd)
1217{
1218 if (nd->flags & LOOKUP_RCU) {
4ad5abb3 1219 int err = inode_permission(nd->inode, MAY_EXEC|MAY_NOT_BLOCK);
52094c8a
AV		 1220 if (err != -ECHILD)
1221 return err;
19660af7 1222 if (unlazy_walk(nd, NULL))
52094c8a
AV		 1223 return -ECHILD;
1224 }
4ad5abb3 1225 return inode_permission(nd->inode, MAY_EXEC);
52094c8a
AV		 1226}
1227
9856fa1b
AV		 1228static inline int handle_dots(struct nameidata *nd, int type)
1229{
1230 if (type == LAST_DOTDOT) {
1231 if (nd->flags & LOOKUP_RCU) {
1232 if (follow_dotdot_rcu(nd))
1233 return -ECHILD;
1234 } else
1235 follow_dotdot(nd);
1236 }
1237 return 0;
1238}
1239
951361f9
AV		 1240static void terminate_walk(struct nameidata *nd)
1241{
1242 if (!(nd->flags & LOOKUP_RCU)) {
1243 path_put(&nd->path);
1244 } else {
1245 nd->flags &= ~LOOKUP_RCU;
5b6ca027
AV		 1246 if (!(nd->flags & LOOKUP_ROOT))
1247 nd->root.mnt = NULL;
951361f9
AV		 1248 rcu_read_unlock();
1249 br_read_unlock(vfsmount_lock);
1250 }
1251}
1252
ce57dfc1
AV		 1253static inline int walk_component(struct nameidata *nd, struct path *path,
1254 struct qstr *name, int type, int follow)
1255{
1256 struct inode *inode;
1257 int err;
1258 /*
1259 * "." and ".." are special - ".." especially so because it has
1260 * to be able to know about the current root directory and
1261 * parent relationships.
1262 */
1263 if (unlikely(type != LAST_NORM))
1264 return handle_dots(nd, type);
1265 err = do_lookup(nd, name, path, &inode);
1266 if (unlikely(err)) {
1267 terminate_walk(nd);
1268 return err;
1269 }
1270 if (!inode) {
1271 path_to_nameidata(path, nd);
1272 terminate_walk(nd);
1273 return -ENOENT;
1274 }
1275 if (unlikely(inode->i_op->follow_link) && follow) {
19660af7
AV		 1276 if (nd->flags & LOOKUP_RCU) {
1277 if (unlikely(unlazy_walk(nd, path->dentry))) {
1278 terminate_walk(nd);
1279 return -ECHILD;
1280 }
1281 }
ce57dfc1
AV		 1282 BUG_ON(inode != path->dentry->d_inode);
1283 return 1;
1284 }
1285 path_to_nameidata(path, nd);
1286 nd->inode = inode;
1287 return 0;
1288}
1289
b356379a
AV		 1290/*
1291 * This limits recursive symlink follows to 8, while
1292 * limiting consecutive symlinks to 40.
1293 *
1294 * Without that kind of total limit, nasty chains of consecutive
1295 * symlinks can cause almost arbitrarily long lookups.
1296 */
1297static inline int nested_symlink(struct path *path, struct nameidata *nd)
1298{
1299 int res;
1300
b356379a
AV		 1301 if (unlikely(current->link_count >= MAX_NESTED_LINKS)) {
1302 path_put_conditional(path, nd);
1303 path_put(&nd->path);
1304 return -ELOOP;
1305 }
1a4022f8 1306 BUG_ON(nd->depth >= MAX_NESTED_LINKS);
b356379a
AV		 1307
1308 nd->depth++;
1309 current->link_count++;
1310
1311 do {
1312 struct path link = *path;
1313 void *cookie;
574197e0
AV		 1314
1315 res = follow_link(&link, nd, &cookie);
b356379a
AV		 1316 if (!res)
1317 res = walk_component(nd, path, &nd->last,
1318 nd->last_type, LOOKUP_FOLLOW);
574197e0 1319 put_link(nd, &link, cookie);
b356379a
AV		 1320 } while (res > 0);
1321
1322 current->link_count--;
1323 nd->depth--;
1324 return res;
1325}
1326
1da177e4
LT		 1327/*
1328 * Name resolution.
ea3834d9
PM		 1329 * This is the basic name resolution function, turning a pathname into
1330 * the final dentry. We expect 'base' to be positive and a directory.
1da177e4 1331 *
ea3834d9
PM		 1332 * Returns 0 and nd will have valid dentry and mnt on success.
1333 * Returns error and drops reference to input namei data on failure.
1da177e4 1334 */
6de88d72 1335static int link_path_walk(const char *name, struct nameidata *nd)
1da177e4
LT		 1336{
1337 struct path next;
1da177e4 1338 int err;
1da177e4
LT		 1339
1340 while (*name=='/')
1341 name++;
1342 if (!*name)
086e183a 1343 return 0;
1da177e4 1344
1da177e4
LT		 1345 /* At this point we know we have a real path component. */
1346 for(;;) {
1347 unsigned long hash;
1348 struct qstr this;
1349 unsigned int c;
fe479a58 1350 int type;
1da177e4 1351
52094c8a 1352 err = may_lookup(nd);
1da177e4
LT		 1353 if (err)
1354 break;
1355
1356 this.name = name;
1357 c = *(const unsigned char *)name;
1358
1359 hash = init_name_hash();
1360 do {
1361 name++;
1362 hash = partial_name_hash(c, hash);
1363 c = *(const unsigned char *)name;
1364 } while (c && (c != '/'));
1365 this.len = name - (const char *) this.name;
1366 this.hash = end_name_hash(hash);
1367
fe479a58
AV		 1368 type = LAST_NORM;
1369 if (this.name[0] == '.') switch (this.len) {
1370 case 2:
16c2cd71 1371 if (this.name[1] == '.') {
fe479a58 1372 type = LAST_DOTDOT;
16c2cd71
AV		 1373 nd->flags |= LOOKUP_JUMPED;
1374 }
fe479a58
AV		 1375 break;
1376 case 1:
1377 type = LAST_DOT;
1378 }
5a202bcd
AV		 1379 if (likely(type == LAST_NORM)) {
1380 struct dentry *parent = nd->path.dentry;
16c2cd71 1381 nd->flags &= ~LOOKUP_JUMPED;
5a202bcd
AV		 1382 if (unlikely(parent->d_flags & DCACHE_OP_HASH)) {
1383 err = parent->d_op->d_hash(parent, nd->inode,
1384 &this);
1385 if (err < 0)
1386 break;
1387 }
1388 }
fe479a58 1389
1da177e4
LT		 1390 /* remove trailing slashes? */
1391 if (!c)
1392 goto last_component;
1393 while (*++name == '/');
1394 if (!*name)
b356379a 1395 goto last_component;
1da177e4 1396
ce57dfc1
AV		 1397 err = walk_component(nd, &next, &this, type, LOOKUP_FOLLOW);
1398 if (err < 0)
1399 return err;
1da177e4 1400
ce57dfc1 1401 if (err) {
b356379a 1402 err = nested_symlink(&next, nd);
1da177e4 1403 if (err)
a7472bab 1404 return err;
31e6b01f 1405 }
1da177e4 1406 err = -ENOTDIR;
31e6b01f 1407 if (!nd->inode->i_op->lookup)
1da177e4
LT		 1408 break;
1409 continue;
1410 /* here ends the main loop */
1411
1da177e4 1412last_component:
b356379a
AV		 1413 nd->last = this;
1414 nd->last_type = type;
086e183a 1415 return 0;
1da177e4 1416 }
951361f9 1417 terminate_walk(nd);
1da177e4
LT		 1418 return err;
1419}
1420
70e9b357
AV		 1421static int path_init(int dfd, const char *name, unsigned int flags,
1422 struct nameidata *nd, struct file **fp)
31e6b01f
NP		 1423{
1424 int retval = 0;
1425 int fput_needed;
1426 struct file *file;
1427
1428 nd->last_type = LAST_ROOT; /* if there are only slashes... */
16c2cd71 1429 nd->flags = flags | LOOKUP_JUMPED;
31e6b01f 1430 nd->depth = 0;
5b6ca027
AV		 1431 if (flags & LOOKUP_ROOT) {
1432 struct inode *inode = nd->root.dentry->d_inode;
73d049a4
AV		 1433 if (*name) {
1434 if (!inode->i_op->lookup)
1435 return -ENOTDIR;
1436 retval = inode_permission(inode, MAY_EXEC);
1437 if (retval)
1438 return retval;
1439 }
5b6ca027
AV		 1440 nd->path = nd->root;
1441 nd->inode = inode;
1442 if (flags & LOOKUP_RCU) {
1443 br_read_lock(vfsmount_lock);
1444 rcu_read_lock();
1445 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1446 } else {
1447 path_get(&nd->path);
1448 }
1449 return 0;
1450 }
1451
31e6b01f 1452 nd->root.mnt = NULL;
31e6b01f
NP		 1453
1454 if (*name=='/') {
e41f7d4e
AV		 1455 if (flags & LOOKUP_RCU) {
1456 br_read_lock(vfsmount_lock);
1457 rcu_read_lock();
1458 set_root_rcu(nd);
1459 } else {
1460 set_root(nd);
1461 path_get(&nd->root);
1462 }
1463 nd->path = nd->root;
31e6b01f 1464 } else if (dfd == AT_FDCWD) {
e41f7d4e
AV		 1465 if (flags & LOOKUP_RCU) {
1466 struct fs_struct *fs = current->fs;
1467 unsigned seq;
31e6b01f 1468
e41f7d4e
AV		 1469 br_read_lock(vfsmount_lock);
1470 rcu_read_lock();
c28cc364 1471
e41f7d4e
AV		 1472 do {
1473 seq = read_seqcount_begin(&fs->seq);
1474 nd->path = fs->pwd;
1475 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1476 } while (read_seqcount_retry(&fs->seq, seq));
1477 } else {
1478 get_fs_pwd(current->fs, &nd->path);
1479 }
31e6b01f
NP		 1480 } else {
1481 struct dentry *dentry;
1482
1abf0c71 1483 file = fget_raw_light(dfd, &fput_needed);
31e6b01f
NP		 1484 retval = -EBADF;
1485 if (!file)
1486 goto out_fail;
1487
1488 dentry = file->f_path.dentry;
1489
f52e0c11
AV		 1490 if (*name) {
1491 retval = -ENOTDIR;
1492 if (!S_ISDIR(dentry->d_inode->i_mode))
1493 goto fput_fail;
31e6b01f 1494
4ad5abb3 1495 retval = inode_permission(dentry->d_inode, MAY_EXEC);
f52e0c11
AV		 1496 if (retval)
1497 goto fput_fail;
1498 }
31e6b01f
NP		 1499
1500 nd->path = file->f_path;
e41f7d4e
AV		 1501 if (flags & LOOKUP_RCU) {
1502 if (fput_needed)
70e9b357 1503 *fp = file;
e41f7d4e
AV		 1504 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1505 br_read_lock(vfsmount_lock);
1506 rcu_read_lock();
1507 } else {
1508 path_get(&file->f_path);
1509 fput_light(file, fput_needed);
1510 }
31e6b01f 1511 }
31e6b01f 1512
31e6b01f 1513 nd->inode = nd->path.dentry->d_inode;
9b4a9b14 1514 return 0;
2dfdd266 1515
9b4a9b14
AV		 1516fput_fail:
1517 fput_light(file, fput_needed);
1518out_fail:
1519 return retval;
1520}
1521
bd92d7fe
AV		 1522static inline int lookup_last(struct nameidata *nd, struct path *path)
1523{
1524 if (nd->last_type == LAST_NORM && nd->last.name[nd->last.len])
1525 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
1526
1527 nd->flags &= ~LOOKUP_PARENT;
1528 return walk_component(nd, path, &nd->last, nd->last_type,
1529 nd->flags & LOOKUP_FOLLOW);
1530}
1531
9b4a9b14 1532/* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
ee0827cd 1533static int path_lookupat(int dfd, const char *name,
9b4a9b14
AV		 1534 unsigned int flags, struct nameidata *nd)
1535{
70e9b357 1536 struct file *base = NULL;
bd92d7fe
AV		 1537 struct path path;
1538 int err;
31e6b01f
NP		 1539
1540 /*
1541 * Path walking is largely split up into 2 different synchronisation
1542 * schemes, rcu-walk and ref-walk (explained in
1543 * Documentation/filesystems/path-lookup.txt). These share much of the
1544 * path walk code, but some things particularly setup, cleanup, and
1545 * following mounts are sufficiently divergent that functions are
1546 * duplicated. Typically there is a function foo(), and its RCU
1547 * analogue, foo_rcu().
1548 *
1549 * -ECHILD is the error number of choice (just to avoid clashes) that
1550 * is returned if some aspect of an rcu-walk fails. Such an error must
1551 * be handled by restarting a traditional ref-walk (which will always
1552 * be able to complete).
1553 */
bd92d7fe 1554 err = path_init(dfd, name, flags | LOOKUP_PARENT, nd, &base);
ee0827cd 1555
bd92d7fe
AV		 1556 if (unlikely(err))
1557 return err;
ee0827cd
AV		 1558
1559 current->total_link_count = 0;
bd92d7fe
AV		 1560 err = link_path_walk(name, nd);
1561
1562 if (!err && !(flags & LOOKUP_PARENT)) {
bd92d7fe
AV		 1563 err = lookup_last(nd, &path);
1564 while (err > 0) {
1565 void *cookie;
1566 struct path link = path;
bd92d7fe 1567 nd->flags |= LOOKUP_PARENT;
574197e0 1568 err = follow_link(&link, nd, &cookie);
bd92d7fe
AV		 1569 if (!err)
1570 err = lookup_last(nd, &path);
574197e0 1571 put_link(nd, &link, cookie);
bd92d7fe
AV		 1572 }
1573 }
ee0827cd 1574
9f1fafee
AV		 1575 if (!err)
1576 err = complete_walk(nd);
bd92d7fe
AV		 1577
1578 if (!err && nd->flags & LOOKUP_DIRECTORY) {
1579 if (!nd->inode->i_op->lookup) {
1580 path_put(&nd->path);
bd23a539 1581 err = -ENOTDIR;
bd92d7fe
AV		 1582 }
1583 }
16c2cd71 1584
70e9b357
AV		 1585 if (base)
1586 fput(base);
ee0827cd 1587
5b6ca027 1588 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) {
2a737871
AV		 1589 path_put(&nd->root);
1590 nd->root.mnt = NULL;
1591 }
bd92d7fe 1592 return err;
ee0827cd 1593}
31e6b01f 1594
ee0827cd
AV		 1595static int do_path_lookup(int dfd, const char *name,
1596 unsigned int flags, struct nameidata *nd)
1597{
1598 int retval = path_lookupat(dfd, name, flags | LOOKUP_RCU, nd);
1599 if (unlikely(retval == -ECHILD))
1600 retval = path_lookupat(dfd, name, flags, nd);
1601 if (unlikely(retval == -ESTALE))
1602 retval = path_lookupat(dfd, name, flags | LOOKUP_REVAL, nd);
31e6b01f
NP		 1603
1604 if (likely(!retval)) {
1605 if (unlikely(!audit_dummy_context())) {
1606 if (nd->path.dentry && nd->inode)
1607 audit_inode(name, nd->path.dentry);
1608 }
1609 }
170aa3d0 1610 return retval;
1da177e4
LT		 1611}
1612
c9c6cac0 1613int kern_path_parent(const char *name, struct nameidata *nd)
5590ff0d 1614{
c9c6cac0 1615 return do_path_lookup(AT_FDCWD, name, LOOKUP_PARENT, nd);
5590ff0d
UD		 1616}
1617
d1811465
AV		 1618int kern_path(const char *name, unsigned int flags, struct path *path)
1619{
1620 struct nameidata nd;
1621 int res = do_path_lookup(AT_FDCWD, name, flags, &nd);
1622 if (!res)
1623 *path = nd.path;
1624 return res;
1625}
1626
16f18200
JJS		 1627/**
1628 * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
1629 * @dentry: pointer to dentry of the base directory
1630 * @mnt: pointer to vfs mount of the base directory
1631 * @name: pointer to file name
1632 * @flags: lookup flags
e0a01249 1633 * @path: pointer to struct path to fill
16f18200
JJS		 1634 */
1635int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt,
1636 const char *name, unsigned int flags,
e0a01249 1637 struct path *path)
16f18200 1638{
e0a01249
AV		 1639 struct nameidata nd;
1640 int err;
1641 nd.root.dentry = dentry;
1642 nd.root.mnt = mnt;
1643 BUG_ON(flags & LOOKUP_PARENT);
5b6ca027 1644 /* the first argument of do_path_lookup() is ignored with LOOKUP_ROOT */
e0a01249
AV		 1645 err = do_path_lookup(AT_FDCWD, name, flags | LOOKUP_ROOT, &nd);
1646 if (!err)
1647 *path = nd.path;
1648 return err;
16f18200
JJS		 1649}
1650
eead1911
CH		 1651static struct dentry *__lookup_hash(struct qstr *name,
1652 struct dentry *base, struct nameidata *nd)
1da177e4 1653{
81fca444 1654 struct inode *inode = base->d_inode;
057f6c01 1655 struct dentry *dentry;
1da177e4
LT		 1656 int err;
1657
4ad5abb3 1658 err = inode_permission(inode, MAY_EXEC);
81fca444
CH		 1659 if (err)
1660 return ERR_PTR(err);
1da177e4 1661
b04f784e
NP		 1662 /*
1663 * Don't bother with __d_lookup: callers are for creat as
1664 * well as unlink, so a lot of the time it would cost
1665 * a double lookup.
6e6b1bd1 1666 */
b04f784e 1667 dentry = d_lookup(base, name);
6e6b1bd1 1668
44396f4b
JB		 1669 if (dentry && d_need_lookup(dentry)) {
1670 /*
1671 * __lookup_hash is called with the parent dir's i_mutex already
1672 * held, so we are good to go here.
1673 */
1674 dentry = d_inode_lookup(base, dentry, nd);
1675 if (IS_ERR(dentry))
1676 return dentry;
1677 }
1678
d2d9e9fb
AV		 1679 if (dentry && (dentry->d_flags & DCACHE_OP_REVALIDATE)) {
1680 int status = d_revalidate(dentry, nd);
1681 if (unlikely(status <= 0)) {
1682 /*
1683 * The dentry failed validation.
1684 * If d_revalidate returned 0 attempt to invalidate
1685 * the dentry otherwise d_revalidate is asking us
1686 * to return a fail status.
1687 */
1688 if (status < 0) {
1689 dput(dentry);
1690 return ERR_PTR(status);
1691 } else if (!d_invalidate(dentry)) {
1692 dput(dentry);
1693 dentry = NULL;
1694 }
1695 }
1696 }
6e6b1bd1 1697
baa03890
NP		 1698 if (!dentry)
1699 dentry = d_alloc_and_lookup(base, name, nd);
5a202bcd 1700
1da177e4
LT		 1701 return dentry;
1702}
1703
057f6c01
JM		 1704/*
1705 * Restricted form of lookup. Doesn't follow links, single-component only,
1706 * needs parent already locked. Doesn't follow mounts.
1707 * SMP-safe.
1708 */
eead1911 1709static struct dentry *lookup_hash(struct nameidata *nd)
057f6c01 1710{
4ac91378 1711 return __lookup_hash(&nd->last, nd->path.dentry, nd);
1da177e4
LT		 1712}
1713
eead1911 1714/**
a6b91919 1715 * lookup_one_len - filesystem helper to lookup single pathname component
eead1911
CH		 1716 * @name: pathname component to lookup
1717 * @base: base directory to lookup from
1718 * @len: maximum length @len should be interpreted to
1719 *
a6b91919
RD		 1720 * Note that this routine is purely a helper for filesystem usage and should
1721 * not be called by generic code. Also note that by using this function the
eead1911
CH		 1722 * nameidata argument is passed to the filesystem methods and a filesystem
1723 * using this helper needs to be prepared for that.
1724 */
057f6c01
JM		 1725struct dentry *lookup_one_len(const char *name, struct dentry *base, int len)
1726{
057f6c01 1727 struct qstr this;
6a96ba54
AV		 1728 unsigned long hash;
1729 unsigned int c;
057f6c01 1730
2f9092e1
DW		 1731 WARN_ON_ONCE(!mutex_is_locked(&base->d_inode->i_mutex));
1732
6a96ba54
AV		 1733 this.name = name;
1734 this.len = len;
1735 if (!len)
1736 return ERR_PTR(-EACCES);
1737
1738 hash = init_name_hash();
1739 while (len--) {
1740 c = *(const unsigned char *)name++;
1741 if (c == '/' || c == '\0')
1742 return ERR_PTR(-EACCES);
1743 hash = partial_name_hash(c, hash);
1744 }
1745 this.hash = end_name_hash(hash);
5a202bcd
AV		 1746 /*
1747 * See if the low-level filesystem might want
1748 * to use its own hash..
1749 */
1750 if (base->d_flags & DCACHE_OP_HASH) {
1751 int err = base->d_op->d_hash(base, base->d_inode, &this);
1752 if (err < 0)
1753 return ERR_PTR(err);
1754 }
eead1911 1755
49705b77 1756 return __lookup_hash(&this, base, NULL);
057f6c01
JM		 1757}
1758
2d8f3038
AV		 1759int user_path_at(int dfd, const char __user *name, unsigned flags,
1760 struct path *path)
1da177e4 1761{
2d8f3038 1762 struct nameidata nd;
f52e0c11 1763 char *tmp = getname_flags(name, flags);
1da177e4 1764 int err = PTR_ERR(tmp);
1da177e4 1765 if (!IS_ERR(tmp)) {
2d8f3038
AV		 1766
1767 BUG_ON(flags & LOOKUP_PARENT);
1768
1769 err = do_path_lookup(dfd, tmp, flags, &nd);
1da177e4 1770 putname(tmp);
2d8f3038
AV		 1771 if (!err)
1772 *path = nd.path;
1da177e4
LT		 1773 }
1774 return err;
1775}
1776
2ad94ae6
AV		 1777static int user_path_parent(int dfd, const char __user *path,
1778 struct nameidata *nd, char **name)
1779{
1780 char *s = getname(path);
1781 int error;
1782
1783 if (IS_ERR(s))
1784 return PTR_ERR(s);
1785
1786 error = do_path_lookup(dfd, s, LOOKUP_PARENT, nd);
1787 if (error)
1788 putname(s);
1789 else
1790 *name = s;
1791
1792 return error;
1793}
1794
1da177e4
LT		 1795/*
1796 * It's inline, so penalty for filesystems that don't use sticky bit is
1797 * minimal.
1798 */
1799static inline int check_sticky(struct inode *dir, struct inode *inode)
1800{
da9592ed
DH		 1801 uid_t fsuid = current_fsuid();
1802
1da177e4
LT		 1803 if (!(dir->i_mode & S_ISVTX))
1804 return 0;
e795b717
SH		 1805 if (current_user_ns() != inode_userns(inode))
1806 goto other_userns;
da9592ed 1807 if (inode->i_uid == fsuid)
1da177e4 1808 return 0;
da9592ed 1809 if (dir->i_uid == fsuid)
1da177e4 1810 return 0;
e795b717
SH		 1811
1812other_userns:
1813 return !ns_capable(inode_userns(inode), CAP_FOWNER);
1da177e4
LT		 1814}
1815
1816/*
1817 * Check whether we can remove a link victim from directory dir, check
1818 * whether the type of victim is right.
1819 * 1. We can't do it if dir is read-only (done in permission())
1820 * 2. We should have write and exec permissions on dir
1821 * 3. We can't remove anything from append-only dir
1822 * 4. We can't do anything with immutable dir (done in permission())
1823 * 5. If the sticky bit on dir is set we should either
1824 * a. be owner of dir, or
1825 * b. be owner of victim, or
1826 * c. have CAP_FOWNER capability
1827 * 6. If the victim is append-only or immutable we can't do antyhing with
1828 * links pointing to it.
1829 * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
1830 * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
1831 * 9. We can't remove a root or mountpoint.
1832 * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
1833 * nfs_async_unlink().
1834 */
858119e1 1835static int may_delete(struct inode *dir,struct dentry *victim,int isdir)
1da177e4
LT		 1836{
1837 int error;
1838
1839 if (!victim->d_inode)
1840 return -ENOENT;
1841
1842 BUG_ON(victim->d_parent->d_inode != dir);
cccc6bba 1843 audit_inode_child(victim, dir);
1da177e4 1844
f419a2e3 1845 error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
1da177e4
LT		 1846 if (error)
1847 return error;
1848 if (IS_APPEND(dir))
1849 return -EPERM;
1850 if (check_sticky(dir, victim->d_inode)||IS_APPEND(victim->d_inode)||
f9454548 1851 IS_IMMUTABLE(victim->d_inode) || IS_SWAPFILE(victim->d_inode))
1da177e4
LT		 1852 return -EPERM;
1853 if (isdir) {
1854 if (!S_ISDIR(victim->d_inode->i_mode))
1855 return -ENOTDIR;
1856 if (IS_ROOT(victim))
1857 return -EBUSY;
1858 } else if (S_ISDIR(victim->d_inode->i_mode))
1859 return -EISDIR;
1860 if (IS_DEADDIR(dir))
1861 return -ENOENT;
1862 if (victim->d_flags & DCACHE_NFSFS_RENAMED)
1863 return -EBUSY;
1864 return 0;
1865}
1866
1867/* Check whether we can create an object with dentry child in directory
1868 * dir.
1869 * 1. We can't do it if child already exists (open has special treatment for
1870 * this case, but since we are inlined it's OK)
1871 * 2. We can't do it if dir is read-only (done in permission())
1872 * 3. We should have write and exec permissions on dir
1873 * 4. We can't do it if dir is immutable (done in permission())
1874 */
a95164d9 1875static inline int may_create(struct inode *dir, struct dentry *child)
1da177e4
LT		 1876{
1877 if (child->d_inode)
1878 return -EEXIST;
1879 if (IS_DEADDIR(dir))
1880 return -ENOENT;
f419a2e3 1881 return inode_permission(dir, MAY_WRITE | MAY_EXEC);
1da177e4
LT		 1882}
1883
1da177e4
LT		 1884/*
1885 * p1 and p2 should be directories on the same fs.
1886 */
1887struct dentry *lock_rename(struct dentry *p1, struct dentry *p2)
1888{
1889 struct dentry *p;
1890
1891 if (p1 == p2) {
f2eace23 1892 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
1da177e4
LT		 1893 return NULL;
1894 }
1895
a11f3a05 1896 mutex_lock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
1da177e4 1897
e2761a11
OH		 1898 p = d_ancestor(p2, p1);
1899 if (p) {
1900 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_PARENT);
1901 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_CHILD);
1902 return p;
1da177e4
LT		 1903 }
1904
e2761a11
OH		 1905 p = d_ancestor(p1, p2);
1906 if (p) {
1907 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
1908 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
1909 return p;
1da177e4
LT		 1910 }
1911
f2eace23
IM		 1912 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
1913 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
1da177e4
LT		 1914 return NULL;
1915}
1916
1917void unlock_rename(struct dentry *p1, struct dentry *p2)
1918{
1b1dcc1b 1919 mutex_unlock(&p1->d_inode->i_mutex);
1da177e4 1920 if (p1 != p2) {
1b1dcc1b 1921 mutex_unlock(&p2->d_inode->i_mutex);
a11f3a05 1922 mutex_unlock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
1da177e4
LT		 1923 }
1924}
1925
1926int vfs_create(struct inode *dir, struct dentry *dentry, int mode,
1927 struct nameidata *nd)
1928{
a95164d9 1929 int error = may_create(dir, dentry);
1da177e4
LT		 1930
1931 if (error)
1932 return error;
1933
acfa4380 1934 if (!dir->i_op->create)
1da177e4
LT		 1935 return -EACCES; /* shouldn't it be ENOSYS? */
1936 mode &= S_IALLUGO;
1937 mode |= S_IFREG;
1938 error = security_inode_create(dir, dentry, mode);
1939 if (error)
1940 return error;
1da177e4 1941 error = dir->i_op->create(dir, dentry, mode, nd);
a74574aa 1942 if (!error)
f38aa942 1943 fsnotify_create(dir, dentry);
1da177e4
LT		 1944 return error;
1945}
1946
73d049a4 1947static int may_open(struct path *path, int acc_mode, int flag)
1da177e4 1948{
3fb64190 1949 struct dentry *dentry = path->dentry;
1da177e4
LT		 1950 struct inode *inode = dentry->d_inode;
1951 int error;
1952
bcda7652
AV		 1953 /* O_PATH? */
1954 if (!acc_mode)
1955 return 0;
1956
1da177e4
LT		 1957 if (!inode)
1958 return -ENOENT;
1959
c8fe8f30
CH		 1960 switch (inode->i_mode & S_IFMT) {
1961 case S_IFLNK:
1da177e4 1962 return -ELOOP;
c8fe8f30
CH		 1963 case S_IFDIR:
1964 if (acc_mode & MAY_WRITE)
1965 return -EISDIR;
1966 break;
1967 case S_IFBLK:
1968 case S_IFCHR:
3fb64190 1969 if (path->mnt->mnt_flags & MNT_NODEV)
1da177e4 1970 return -EACCES;
c8fe8f30
CH		 1971 /*FALLTHRU*/
1972 case S_IFIFO:
1973 case S_IFSOCK:
1da177e4 1974 flag &= ~O_TRUNC;
c8fe8f30 1975 break;
4a3fd211 1976 }
b41572e9 1977
3fb64190 1978 error = inode_permission(inode, acc_mode);
b41572e9
DH		 1979 if (error)
1980 return error;
6146f0d5 1981
1da177e4
LT		 1982 /*
1983 * An append-only file must be opened in append mode for writing.
1984 */
1985 if (IS_APPEND(inode)) {
8737c930 1986 if ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND))
7715b521 1987 return -EPERM;
1da177e4 1988 if (flag & O_TRUNC)
7715b521 1989 return -EPERM;
1da177e4
LT		 1990 }
1991
1992 /* O_NOATIME can only be set by the owner or superuser */
2e149670 1993 if (flag & O_NOATIME && !inode_owner_or_capable(inode))
7715b521 1994 return -EPERM;
1da177e4
LT		 1995
1996 /*
1997 * Ensure there are no outstanding leases on the file.
1998 */
b65a9cfc 1999 return break_lease(inode, flag);
7715b521 2000}
1da177e4 2001
e1181ee6 2002static int handle_truncate(struct file *filp)
7715b521 2003{
e1181ee6 2004 struct path *path = &filp->f_path;
7715b521
AV		 2005 struct inode *inode = path->dentry->d_inode;
2006 int error = get_write_access(inode);
2007 if (error)
2008 return error;
2009 /*
2010 * Refuse to truncate files with mandatory locks held on them.
2011 */
2012 error = locks_verify_locked(inode);
2013 if (!error)
ea0d3ab2 2014 error = security_path_truncate(path);
7715b521
AV		 2015 if (!error) {
2016 error = do_truncate(path->dentry, 0,
2017 ATTR_MTIME|ATTR_CTIME|ATTR_OPEN,
e1181ee6 2018 filp);
7715b521
AV		 2019 }
2020 put_write_access(inode);
acd0c935 2021 return error;
1da177e4
LT		 2022}
2023
d57999e1
DH		 2024static inline int open_to_namei_flags(int flag)
2025{
8a5e929d
AV		 2026 if ((flag & O_ACCMODE) == 3)
2027 flag--;
d57999e1
DH		 2028 return flag;
2029}
2030
31e6b01f 2031/*
fe2d35ff 2032 * Handle the last step of open()
31e6b01f 2033 */
fb1cc555 2034static struct file *do_last(struct nameidata *nd, struct path *path,
c3e380b0 2035 const struct open_flags *op, const char *pathname)
fb1cc555 2036{
a1e28038 2037 struct dentry *dir = nd->path.dentry;
6c0d46c4 2038 struct dentry *dentry;
ca344a89 2039 int open_flag = op->open_flag;
6c0d46c4 2040 int will_truncate = open_flag & O_TRUNC;
ca344a89 2041 int want_write = 0;
bcda7652 2042 int acc_mode = op->acc_mode;
fb1cc555 2043 struct file *filp;
16c2cd71 2044 int error;
1f36f774 2045
c3e380b0
AV		 2046 nd->flags &= ~LOOKUP_PARENT;
2047 nd->flags |= op->intent;
2048
1f36f774
AV		 2049 switch (nd->last_type) {
2050 case LAST_DOTDOT:
176306f5 2051 case LAST_DOT:
fe2d35ff
AV		 2052 error = handle_dots(nd, nd->last_type);
2053 if (error)
2054 return ERR_PTR(error);
1f36f774 2055 /* fallthrough */
1f36f774 2056 case LAST_ROOT:
9f1fafee 2057 error = complete_walk(nd);
16c2cd71 2058 if (error)
9f1fafee 2059 return ERR_PTR(error);
fe2d35ff 2060 audit_inode(pathname, nd->path.dentry);
ca344a89 2061 if (open_flag & O_CREAT) {
fe2d35ff
AV		 2062 error = -EISDIR;
2063 goto exit;
2064 }
2065 goto ok;
1f36f774 2066 case LAST_BIND:
9f1fafee 2067 error = complete_walk(nd);
16c2cd71 2068 if (error)
9f1fafee 2069 return ERR_PTR(error);
1f36f774 2070 audit_inode(pathname, dir);
67ee3ad2 2071 goto ok;
1f36f774 2072 }
67ee3ad2 2073
ca344a89 2074 if (!(open_flag & O_CREAT)) {
bcda7652 2075 int symlink_ok = 0;
fe2d35ff
AV		 2076 if (nd->last.name[nd->last.len])
2077 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
bcda7652
AV		 2078 if (open_flag & O_PATH && !(nd->flags & LOOKUP_FOLLOW))
2079 symlink_ok = 1;
fe2d35ff 2080 /* we _can_ be in RCU mode here */
ce57dfc1
AV		 2081 error = walk_component(nd, path, &nd->last, LAST_NORM,
2082 !symlink_ok);
2083 if (error < 0)
fe2d35ff 2084 return ERR_PTR(error);
ce57dfc1 2085 if (error) /* symlink */
fe2d35ff 2086 return NULL;
fe2d35ff 2087 /* sayonara */
9f1fafee
AV		 2088 error = complete_walk(nd);
2089 if (error)
2090 return ERR_PTR(-ECHILD);
fe2d35ff
AV		 2091
2092 error = -ENOTDIR;
2093 if (nd->flags & LOOKUP_DIRECTORY) {
ce57dfc1 2094 if (!nd->inode->i_op->lookup)
fe2d35ff
AV		 2095 goto exit;
2096 }
2097 audit_inode(pathname, nd->path.dentry);
2098 goto ok;
2099 }
2100
2101 /* create side of things */
9f1fafee
AV		 2102 error = complete_walk(nd);
2103 if (error)
2104 return ERR_PTR(error);
fe2d35ff
AV		 2105
2106 audit_inode(pathname, dir);
16c2cd71 2107 error = -EISDIR;
1f36f774 2108 /* trailing slashes? */
31e6b01f
NP		 2109 if (nd->last.name[nd->last.len])
2110 goto exit;
a2c36b45 2111
a1e28038
AV		 2112 mutex_lock(&dir->d_inode->i_mutex);
2113
6c0d46c4
AV		 2114 dentry = lookup_hash(nd);
2115 error = PTR_ERR(dentry);
2116 if (IS_ERR(dentry)) {
fb1cc555
AV		 2117 mutex_unlock(&dir->d_inode->i_mutex);
2118 goto exit;
2119 }
2120
6c0d46c4
AV		 2121 path->dentry = dentry;
2122 path->mnt = nd->path.mnt;
2123
fb1cc555 2124 /* Negative dentry, just create the file */
6c0d46c4
AV		 2125 if (!dentry->d_inode) {
2126 int mode = op->mode;
2127 if (!IS_POSIXACL(dir->d_inode))
2128 mode &= ~current_umask();
fb1cc555
AV		 2129 /*
2130 * This write is needed to ensure that a
6c0d46c4 2131 * rw->ro transition does not occur between
fb1cc555
AV		 2132 * the time when the file is created and when
2133 * a permanent write count is taken through
2134 * the 'struct file' in nameidata_to_filp().
2135 */
2136 error = mnt_want_write(nd->path.mnt);
2137 if (error)
2138 goto exit_mutex_unlock;
ca344a89 2139 want_write = 1;
9b44f1b3 2140 /* Don't check for write permission, don't truncate */
ca344a89 2141 open_flag &= ~O_TRUNC;
6c0d46c4 2142 will_truncate = 0;
bcda7652 2143 acc_mode = MAY_OPEN;
6c0d46c4
AV		 2144 error = security_path_mknod(&nd->path, dentry, mode, 0);
2145 if (error)
2146 goto exit_mutex_unlock;
2147 error = vfs_create(dir->d_inode, dentry, mode, nd);
2148 if (error)
2149 goto exit_mutex_unlock;
2150 mutex_unlock(&dir->d_inode->i_mutex);
2151 dput(nd->path.dentry);
2152 nd->path.dentry = dentry;
ca344a89 2153 goto common;
fb1cc555
AV		 2154 }
2155
2156 /*
2157 * It already exists.
2158 */
2159 mutex_unlock(&dir->d_inode->i_mutex);
2160 audit_inode(pathname, path->dentry);
2161
2162 error = -EEXIST;
ca344a89 2163 if (open_flag & O_EXCL)
fb1cc555
AV		 2164 goto exit_dput;
2165
9875cf80
DH		 2166 error = follow_managed(path, nd->flags);
2167 if (error < 0)
2168 goto exit_dput;
fb1cc555
AV		 2169
2170 error = -ENOENT;
2171 if (!path->dentry->d_inode)
2172 goto exit_dput;
9e67f361
AV		 2173
2174 if (path->dentry->d_inode->i_op->follow_link)
fb1cc555 2175 return NULL;
fb1cc555
AV		 2176
2177 path_to_nameidata(path, nd);
31e6b01f 2178 nd->inode = path->dentry->d_inode;
fb1cc555 2179 error = -EISDIR;
31e6b01f 2180 if (S_ISDIR(nd->inode->i_mode))
fb1cc555 2181 goto exit;
67ee3ad2 2182ok:
6c0d46c4
AV		 2183 if (!S_ISREG(nd->inode->i_mode))
2184 will_truncate = 0;
2185
0f9d1a10
AV		 2186 if (will_truncate) {
2187 error = mnt_want_write(nd->path.mnt);
2188 if (error)
2189 goto exit;
ca344a89 2190 want_write = 1;
0f9d1a10 2191 }
ca344a89 2192common:
bcda7652 2193 error = may_open(&nd->path, acc_mode, open_flag);
ca344a89 2194 if (error)
0f9d1a10 2195 goto exit;
0f9d1a10
AV		 2196 filp = nameidata_to_filp(nd);
2197 if (!IS_ERR(filp)) {
2198 error = ima_file_check(filp, op->acc_mode);
2199 if (error) {
2200 fput(filp);
2201 filp = ERR_PTR(error);
2202 }
2203 }
2204 if (!IS_ERR(filp)) {
2205 if (will_truncate) {
2206 error = handle_truncate(filp);
2207 if (error) {
2208 fput(filp);
2209 filp = ERR_PTR(error);
2210 }
2211 }
2212 }
ca344a89
AV		 2213out:
2214 if (want_write)
0f9d1a10
AV		 2215 mnt_drop_write(nd->path.mnt);
2216 path_put(&nd->path);
fb1cc555
AV		 2217 return filp;
2218
2219exit_mutex_unlock:
2220 mutex_unlock(&dir->d_inode->i_mutex);
2221exit_dput:
2222 path_put_conditional(path, nd);
2223exit:
ca344a89
AV		 2224 filp = ERR_PTR(error);
2225 goto out;
fb1cc555
AV		 2226}
2227
13aab428 2228static struct file *path_openat(int dfd, const char *pathname,
73d049a4 2229 struct nameidata *nd, const struct open_flags *op, int flags)
1da177e4 2230{
fe2d35ff 2231 struct file *base = NULL;
4a3fd211 2232 struct file *filp;
9850c056 2233 struct path path;
13aab428 2234 int error;
31e6b01f
NP		 2235
2236 filp = get_empty_filp();
2237 if (!filp)
2238 return ERR_PTR(-ENFILE);
2239
47c805dc 2240 filp->f_flags = op->open_flag;
73d049a4
AV		 2241 nd->intent.open.file = filp;
2242 nd->intent.open.flags = open_to_namei_flags(op->open_flag);
2243 nd->intent.open.create_mode = op->mode;
31e6b01f 2244
73d049a4 2245 error = path_init(dfd, pathname, flags | LOOKUP_PARENT, nd, &base);
31e6b01f 2246 if (unlikely(error))
13aab428 2247 goto out_filp;
31e6b01f 2248
fe2d35ff 2249 current->total_link_count = 0;
73d049a4 2250 error = link_path_walk(pathname, nd);
31e6b01f
NP		 2251 if (unlikely(error))
2252 goto out_filp;
1da177e4 2253
73d049a4 2254 filp = do_last(nd, &path, op, pathname);
806b681c 2255 while (unlikely(!filp)) { /* trailing symlink */
7b9337aa 2256 struct path link = path;
def4af30 2257 void *cookie;
574197e0 2258 if (!(nd->flags & LOOKUP_FOLLOW)) {
73d049a4
AV		 2259 path_put_conditional(&path, nd);
2260 path_put(&nd->path);
40b39136
AV		 2261 filp = ERR_PTR(-ELOOP);
2262 break;
2263 }
73d049a4
AV		 2264 nd->flags |= LOOKUP_PARENT;
2265 nd->flags &= ~(LOOKUP_OPEN|LOOKUP_CREATE|LOOKUP_EXCL);
574197e0 2266 error = follow_link(&link, nd, &cookie);
c3e380b0 2267 if (unlikely(error))
f1afe9ef 2268 filp = ERR_PTR(error);
c3e380b0 2269 else
73d049a4 2270 filp = do_last(nd, &path, op, pathname);
574197e0 2271 put_link(nd, &link, cookie);
806b681c 2272 }
10fa8e62 2273out:
73d049a4
AV		 2274 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT))
2275 path_put(&nd->root);
fe2d35ff
AV		 2276 if (base)
2277 fput(base);
73d049a4 2278 release_open_intent(nd);
10fa8e62 2279 return filp;
1da177e4 2280
31e6b01f 2281out_filp:
806b681c 2282 filp = ERR_PTR(error);
10fa8e62 2283 goto out;
1da177e4
LT		 2284}
2285
13aab428
AV		 2286struct file *do_filp_open(int dfd, const char *pathname,
2287 const struct open_flags *op, int flags)
2288{
73d049a4 2289 struct nameidata nd;
13aab428
AV		 2290 struct file *filp;
2291
73d049a4 2292 filp = path_openat(dfd, pathname, &nd, op, flags | LOOKUP_RCU);
13aab428 2293 if (unlikely(filp == ERR_PTR(-ECHILD)))
73d049a4 2294 filp = path_openat(dfd, pathname, &nd, op, flags);
13aab428 2295 if (unlikely(filp == ERR_PTR(-ESTALE)))
73d049a4 2296 filp = path_openat(dfd, pathname, &nd, op, flags | LOOKUP_REVAL);
13aab428
AV		 2297 return filp;
2298}
2299
73d049a4
AV		 2300struct file *do_file_open_root(struct dentry *dentry, struct vfsmount *mnt,
2301 const char *name, const struct open_flags *op, int flags)
2302{
2303 struct nameidata nd;
2304 struct file *file;
2305
2306 nd.root.mnt = mnt;
2307 nd.root.dentry = dentry;
2308
2309 flags |= LOOKUP_ROOT;
2310
bcda7652 2311 if (dentry->d_inode->i_op->follow_link && op->intent & LOOKUP_OPEN)
73d049a4
AV		 2312 return ERR_PTR(-ELOOP);
2313
2314 file = path_openat(-1, name, &nd, op, flags | LOOKUP_RCU);
2315 if (unlikely(file == ERR_PTR(-ECHILD)))
2316 file = path_openat(-1, name, &nd, op, flags);
2317 if (unlikely(file == ERR_PTR(-ESTALE)))
2318 file = path_openat(-1, name, &nd, op, flags | LOOKUP_REVAL);
2319 return file;
2320}
2321
ed75e95d 2322struct dentry *kern_path_create(int dfd, const char *pathname, struct path *path, int is_dir)
1da177e4 2323{
c663e5d8 2324 struct dentry *dentry = ERR_PTR(-EEXIST);
ed75e95d
AV		 2325 struct nameidata nd;
2326 int error = do_path_lookup(dfd, pathname, LOOKUP_PARENT, &nd);
2327 if (error)
2328 return ERR_PTR(error);
1da177e4 2329
c663e5d8
CH		 2330 /*
2331 * Yucky last component or no last component at all?
2332 * (foo/., foo/.., /////)
2333 */
ed75e95d
AV		 2334 if (nd.last_type != LAST_NORM)
2335 goto out;
2336 nd.flags &= ~LOOKUP_PARENT;
2337 nd.flags |= LOOKUP_CREATE | LOOKUP_EXCL;
2338 nd.intent.open.flags = O_EXCL;
c663e5d8
CH		 2339
2340 /*
2341 * Do the final lookup.
2342 */
ed75e95d
AV		 2343 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
2344 dentry = lookup_hash(&nd);
1da177e4
LT		 2345 if (IS_ERR(dentry))
2346 goto fail;
c663e5d8 2347
e9baf6e5
AV		 2348 if (dentry->d_inode)
2349 goto eexist;
c663e5d8
CH		 2350 /*
2351 * Special case - lookup gave negative, but... we had foo/bar/
2352 * From the vfs_mknod() POV we just have a negative dentry -
2353 * all is fine. Let's be bastards - you had / on the end, you've
2354 * been asking for (non-existent) directory. -ENOENT for you.
2355 */
ed75e95d 2356 if (unlikely(!is_dir && nd.last.name[nd.last.len])) {
e9baf6e5
AV		 2357 dput(dentry);
2358 dentry = ERR_PTR(-ENOENT);
ed75e95d 2359 goto fail;
e9baf6e5 2360 }
ed75e95d 2361 *path = nd.path;
1da177e4 2362 return dentry;
e9baf6e5 2363eexist:
1da177e4 2364 dput(dentry);
e9baf6e5 2365 dentry = ERR_PTR(-EEXIST);
1da177e4 2366fail:
ed75e95d
AV		 2367 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2368out:
2369 path_put(&nd.path);
1da177e4
LT		 2370 return dentry;
2371}
dae6ad8f
AV		 2372EXPORT_SYMBOL(kern_path_create);
2373
2374struct dentry *user_path_create(int dfd, const char __user *pathname, struct path *path, int is_dir)
2375{
2376 char *tmp = getname(pathname);
2377 struct dentry *res;
2378 if (IS_ERR(tmp))
2379 return ERR_CAST(tmp);
2380 res = kern_path_create(dfd, tmp, path, is_dir);
2381 putname(tmp);
2382 return res;
2383}
2384EXPORT_SYMBOL(user_path_create);
2385
1da177e4
LT		 2386int vfs_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t dev)
2387{
a95164d9 2388 int error = may_create(dir, dentry);
1da177e4
LT		 2389
2390 if (error)
2391 return error;
2392
e795b717
SH		 2393 if ((S_ISCHR(mode) || S_ISBLK(mode)) &&
2394 !ns_capable(inode_userns(dir), CAP_MKNOD))
1da177e4
LT		 2395 return -EPERM;
2396
acfa4380 2397 if (!dir->i_op->mknod)
1da177e4
LT		 2398 return -EPERM;
2399
08ce5f16
SH		 2400 error = devcgroup_inode_mknod(mode, dev);
2401 if (error)
2402 return error;
2403
1da177e4
LT		 2404 error = security_inode_mknod(dir, dentry, mode, dev);
2405 if (error)
2406 return error;
2407
1da177e4 2408 error = dir->i_op->mknod(dir, dentry, mode, dev);
a74574aa 2409 if (!error)
f38aa942 2410 fsnotify_create(dir, dentry);
1da177e4
LT		 2411 return error;
2412}
2413
463c3197
DH		 2414static int may_mknod(mode_t mode)
2415{
2416 switch (mode & S_IFMT) {
2417 case S_IFREG:
2418 case S_IFCHR:
2419 case S_IFBLK:
2420 case S_IFIFO:
2421 case S_IFSOCK:
2422 case 0: /* zero mode translates to S_IFREG */
2423 return 0;
2424 case S_IFDIR:
2425 return -EPERM;
2426 default:
2427 return -EINVAL;
2428 }
2429}
2430
2e4d0924
HC		 2431SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, int, mode,
2432 unsigned, dev)
1da177e4 2433{
2ad94ae6 2434 struct dentry *dentry;
dae6ad8f
AV		 2435 struct path path;
2436 int error;
1da177e4
LT		 2437
2438 if (S_ISDIR(mode))
2439 return -EPERM;
1da177e4 2440
dae6ad8f
AV		 2441 dentry = user_path_create(dfd, filename, &path, 0);
2442 if (IS_ERR(dentry))
2443 return PTR_ERR(dentry);
2ad94ae6 2444
dae6ad8f 2445 if (!IS_POSIXACL(path.dentry->d_inode))
ce3b0f8d 2446 mode &= ~current_umask();
463c3197
DH		 2447 error = may_mknod(mode);
2448 if (error)
2449 goto out_dput;
dae6ad8f 2450 error = mnt_want_write(path.mnt);
463c3197
DH		 2451 if (error)
2452 goto out_dput;
dae6ad8f 2453 error = security_path_mknod(&path, dentry, mode, dev);
be6d3e56
KT		 2454 if (error)
2455 goto out_drop_write;
463c3197 2456 switch (mode & S_IFMT) {
1da177e4 2457 case 0: case S_IFREG:
dae6ad8f 2458 error = vfs_create(path.dentry->d_inode,dentry,mode,NULL);
1da177e4
LT		 2459 break;
2460 case S_IFCHR: case S_IFBLK:
dae6ad8f 2461 error = vfs_mknod(path.dentry->d_inode,dentry,mode,
1da177e4
LT		 2462 new_decode_dev(dev));
2463 break;
2464 case S_IFIFO: case S_IFSOCK:
dae6ad8f 2465 error = vfs_mknod(path.dentry->d_inode,dentry,mode,0);
1da177e4 2466 break;
1da177e4 2467 }
be6d3e56 2468out_drop_write:
dae6ad8f 2469 mnt_drop_write(path.mnt);
463c3197
DH		 2470out_dput:
2471 dput(dentry);
dae6ad8f
AV		 2472 mutex_unlock(&path.dentry->d_inode->i_mutex);
2473 path_put(&path);
1da177e4
LT		 2474
2475 return error;
2476}
2477
3480b257 2478SYSCALL_DEFINE3(mknod, const char __user *, filename, int, mode, unsigned, dev)
5590ff0d
UD		 2479{
2480 return sys_mknodat(AT_FDCWD, filename, mode, dev);
2481}
2482
1da177e4
LT		 2483int vfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
2484{
a95164d9 2485 int error = may_create(dir, dentry);
1da177e4
LT		 2486
2487 if (error)
2488 return error;
2489
acfa4380 2490 if (!dir->i_op->mkdir)
1da177e4
LT		 2491 return -EPERM;
2492
2493 mode &= (S_IRWXUGO|S_ISVTX);
2494 error = security_inode_mkdir(dir, dentry, mode);
2495 if (error)
2496 return error;
2497
1da177e4 2498 error = dir->i_op->mkdir(dir, dentry, mode);
a74574aa 2499 if (!error)
f38aa942 2500 fsnotify_mkdir(dir, dentry);
1da177e4
LT		 2501 return error;
2502}
2503
2e4d0924 2504SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, int, mode)
1da177e4 2505{
6902d925 2506 struct dentry *dentry;
dae6ad8f
AV		 2507 struct path path;
2508 int error;
1da177e4 2509
dae6ad8f 2510 dentry = user_path_create(dfd, pathname, &path, 1);
6902d925 2511 if (IS_ERR(dentry))
dae6ad8f 2512 return PTR_ERR(dentry);
1da177e4 2513
dae6ad8f 2514 if (!IS_POSIXACL(path.dentry->d_inode))
ce3b0f8d 2515 mode &= ~current_umask();
dae6ad8f 2516 error = mnt_want_write(path.mnt);
463c3197
DH		 2517 if (error)
2518 goto out_dput;
dae6ad8f 2519 error = security_path_mkdir(&path, dentry, mode);
be6d3e56
KT		 2520 if (error)
2521 goto out_drop_write;
dae6ad8f 2522 error = vfs_mkdir(path.dentry->d_inode, dentry, mode);
be6d3e56 2523out_drop_write:
dae6ad8f 2524 mnt_drop_write(path.mnt);
463c3197 2525out_dput:
6902d925 2526 dput(dentry);
dae6ad8f
AV		 2527 mutex_unlock(&path.dentry->d_inode->i_mutex);
2528 path_put(&path);
1da177e4
LT		 2529 return error;
2530}
2531
3cdad428 2532SYSCALL_DEFINE2(mkdir, const char __user *, pathname, int, mode)
5590ff0d
UD		 2533{
2534 return sys_mkdirat(AT_FDCWD, pathname, mode);
2535}
2536
1da177e4 2537/*
a71905f0
SW		 2538 * The dentry_unhash() helper will try to drop the dentry early: we
2539 * should have a usage count of 2 if we're the only user of this
2540 * dentry, and if that is true (possibly after pruning the dcache),
2541 * then we drop the dentry now.
1da177e4
LT		 2542 *
2543 * A low-level filesystem can, if it choses, legally
2544 * do a
2545 *
2546 * if (!d_unhashed(dentry))
2547 * return -EBUSY;
2548 *
2549 * if it cannot handle the case of removing a directory
2550 * that is still in use by something else..
2551 */
2552void dentry_unhash(struct dentry *dentry)
2553{
dc168427 2554 shrink_dcache_parent(dentry);
1da177e4 2555 spin_lock(&dentry->d_lock);
64252c75 2556 if (dentry->d_count == 1)
1da177e4
LT		 2557 __d_drop(dentry);
2558 spin_unlock(&dentry->d_lock);
1da177e4
LT		 2559}
2560
2561int vfs_rmdir(struct inode *dir, struct dentry *dentry)
2562{
2563 int error = may_delete(dir, dentry, 1);
2564
2565 if (error)
2566 return error;
2567
acfa4380 2568 if (!dir->i_op->rmdir)
1da177e4
LT		 2569 return -EPERM;
2570
1b1dcc1b 2571 mutex_lock(&dentry->d_inode->i_mutex);
912dbc15
SW		 2572
2573 error = -EBUSY;
1da177e4 2574 if (d_mountpoint(dentry))
912dbc15
SW		 2575 goto out;
2576
2577 error = security_inode_rmdir(dir, dentry);
2578 if (error)
2579 goto out;
2580
3cebde24 2581 shrink_dcache_parent(dentry);
912dbc15
SW		 2582 error = dir->i_op->rmdir(dir, dentry);
2583 if (error)
2584 goto out;
2585
2586 dentry->d_inode->i_flags |= S_DEAD;
2587 dont_mount(dentry);
2588
2589out:
1b1dcc1b 2590 mutex_unlock(&dentry->d_inode->i_mutex);
912dbc15 2591 if (!error)
1da177e4 2592 d_delete(dentry);
1da177e4
LT		 2593 return error;
2594}
2595
5590ff0d 2596static long do_rmdir(int dfd, const char __user *pathname)
1da177e4
LT		 2597{
2598 int error = 0;
2599 char * name;
2600 struct dentry *dentry;
2601 struct nameidata nd;
2602
2ad94ae6 2603 error = user_path_parent(dfd, pathname, &nd, &name);
1da177e4 2604 if (error)
2ad94ae6 2605 return error;
1da177e4
LT		 2606
2607 switch(nd.last_type) {
0612d9fb
OH		 2608 case LAST_DOTDOT:
2609 error = -ENOTEMPTY;
2610 goto exit1;
2611 case LAST_DOT:
2612 error = -EINVAL;
2613 goto exit1;
2614 case LAST_ROOT:
2615 error = -EBUSY;
2616 goto exit1;
1da177e4 2617 }
0612d9fb
OH		 2618
2619 nd.flags &= ~LOOKUP_PARENT;
2620
4ac91378 2621 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
49705b77 2622 dentry = lookup_hash(&nd);
1da177e4 2623 error = PTR_ERR(dentry);
6902d925
DH		 2624 if (IS_ERR(dentry))
2625 goto exit2;
e6bc45d6
TT		 2626 if (!dentry->d_inode) {
2627 error = -ENOENT;
2628 goto exit3;
2629 }
0622753b
DH		 2630 error = mnt_want_write(nd.path.mnt);
2631 if (error)
2632 goto exit3;
be6d3e56
KT		 2633 error = security_path_rmdir(&nd.path, dentry);
2634 if (error)
2635 goto exit4;
4ac91378 2636 error = vfs_rmdir(nd.path.dentry->d_inode, dentry);
be6d3e56 2637exit4:
0622753b
DH		 2638 mnt_drop_write(nd.path.mnt);
2639exit3:
6902d925
DH		 2640 dput(dentry);
2641exit2:
4ac91378 2642 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
1da177e4 2643exit1:
1d957f9b 2644 path_put(&nd.path);
1da177e4
LT		 2645 putname(name);
2646 return error;
2647}
2648
3cdad428 2649SYSCALL_DEFINE1(rmdir, const char __user *, pathname)
5590ff0d
UD		 2650{
2651 return do_rmdir(AT_FDCWD, pathname);
2652}
2653
1da177e4
LT		 2654int vfs_unlink(struct inode *dir, struct dentry *dentry)
2655{
2656 int error = may_delete(dir, dentry, 0);
2657
2658 if (error)
2659 return error;
2660
acfa4380 2661 if (!dir->i_op->unlink)
1da177e4
LT		 2662 return -EPERM;
2663
1b1dcc1b 2664 mutex_lock(&dentry->d_inode->i_mutex);
1da177e4
LT		 2665 if (d_mountpoint(dentry))
2666 error = -EBUSY;
2667 else {
2668 error = security_inode_unlink(dir, dentry);
bec1052e 2669 if (!error) {
1da177e4 2670 error = dir->i_op->unlink(dir, dentry);
bec1052e 2671 if (!error)
d83c49f3 2672 dont_mount(dentry);
bec1052e 2673 }
1da177e4 2674 }
1b1dcc1b 2675 mutex_unlock(&dentry->d_inode->i_mutex);
1da177e4
LT		 2676
2677 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
2678 if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) {
ece95912 2679 fsnotify_link_count(dentry->d_inode);
e234f35c 2680 d_delete(dentry);
1da177e4 2681 }
0eeca283 2682
1da177e4
LT		 2683 return error;
2684}
2685
2686/*
2687 * Make sure that the actual truncation of the file will occur outside its
1b1dcc1b 2688 * directory's i_mutex. Truncate can take a long time if there is a lot of
1da177e4
LT		 2689 * writeout happening, and we don't want to prevent access to the directory
2690 * while waiting on the I/O.
2691 */
5590ff0d 2692static long do_unlinkat(int dfd, const char __user *pathname)
1da177e4 2693{
2ad94ae6
AV		 2694 int error;
2695 char *name;
1da177e4
LT		 2696 struct dentry *dentry;
2697 struct nameidata nd;
2698 struct inode *inode = NULL;
2699
2ad94ae6 2700 error = user_path_parent(dfd, pathname, &nd, &name);
1da177e4 2701 if (error)
2ad94ae6
AV		 2702 return error;
2703
1da177e4
LT		 2704 error = -EISDIR;
2705 if (nd.last_type != LAST_NORM)
2706 goto exit1;
0612d9fb
OH		 2707
2708 nd.flags &= ~LOOKUP_PARENT;
2709
4ac91378 2710 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
49705b77 2711 dentry = lookup_hash(&nd);
1da177e4
LT		 2712 error = PTR_ERR(dentry);
2713 if (!IS_ERR(dentry)) {
2714 /* Why not before? Because we want correct error value */
50338b88
TE		 2715 if (nd.last.name[nd.last.len])
2716 goto slashes;
1da177e4 2717 inode = dentry->d_inode;
50338b88 2718 if (!inode)
e6bc45d6
TT		 2719 goto slashes;
2720 ihold(inode);
0622753b
DH		 2721 error = mnt_want_write(nd.path.mnt);
2722 if (error)
2723 goto exit2;
be6d3e56
KT		 2724 error = security_path_unlink(&nd.path, dentry);
2725 if (error)
2726 goto exit3;
4ac91378 2727 error = vfs_unlink(nd.path.dentry->d_inode, dentry);
be6d3e56 2728exit3:
0622753b 2729 mnt_drop_write(nd.path.mnt);
1da177e4
LT		 2730 exit2:
2731 dput(dentry);
2732 }
4ac91378 2733 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
1da177e4
LT		 2734 if (inode)
2735 iput(inode); /* truncate the inode here */
2736exit1:
1d957f9b 2737 path_put(&nd.path);
1da177e4
LT		 2738 putname(name);
2739 return error;
2740
2741slashes:
2742 error = !dentry->d_inode ? -ENOENT :
2743 S_ISDIR(dentry->d_inode->i_mode) ? -EISDIR : -ENOTDIR;
2744 goto exit2;
2745}
2746
2e4d0924 2747SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag)
5590ff0d
UD		 2748{
2749 if ((flag & ~AT_REMOVEDIR) != 0)
2750 return -EINVAL;
2751
2752 if (flag & AT_REMOVEDIR)
2753 return do_rmdir(dfd, pathname);
2754
2755 return do_unlinkat(dfd, pathname);
2756}
2757
3480b257 2758SYSCALL_DEFINE1(unlink, const char __user *, pathname)
5590ff0d
UD		 2759{
2760 return do_unlinkat(AT_FDCWD, pathname);
2761}
2762
db2e747b 2763int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname)
1da177e4 2764{
a95164d9 2765 int error = may_create(dir, dentry);
1da177e4
LT		 2766
2767 if (error)
2768 return error;
2769
acfa4380 2770 if (!dir->i_op->symlink)
1da177e4
LT		 2771 return -EPERM;
2772
2773 error = security_inode_symlink(dir, dentry, oldname);
2774 if (error)
2775 return error;
2776
1da177e4 2777 error = dir->i_op->symlink(dir, dentry, oldname);
a74574aa 2778 if (!error)
f38aa942 2779 fsnotify_create(dir, dentry);
1da177e4
LT		 2780 return error;
2781}
2782
2e4d0924
HC		 2783SYSCALL_DEFINE3(symlinkat, const char __user *, oldname,
2784 int, newdfd, const char __user *, newname)
1da177e4 2785{
2ad94ae6
AV		 2786 int error;
2787 char *from;
6902d925 2788 struct dentry *dentry;
dae6ad8f 2789 struct path path;
1da177e4
LT		 2790
2791 from = getname(oldname);
2ad94ae6 2792 if (IS_ERR(from))
1da177e4 2793 return PTR_ERR(from);
1da177e4 2794
dae6ad8f 2795 dentry = user_path_create(newdfd, newname, &path, 0);
6902d925
DH		 2796 error = PTR_ERR(dentry);
2797 if (IS_ERR(dentry))
dae6ad8f 2798 goto out_putname;
6902d925 2799
dae6ad8f 2800 error = mnt_want_write(path.mnt);
75c3f29d
DH		 2801 if (error)
2802 goto out_dput;
dae6ad8f 2803 error = security_path_symlink(&path, dentry, from);
be6d3e56
KT		 2804 if (error)
2805 goto out_drop_write;
dae6ad8f 2806 error = vfs_symlink(path.dentry->d_inode, dentry, from);
be6d3e56 2807out_drop_write:
dae6ad8f 2808 mnt_drop_write(path.mnt);
75c3f29d 2809out_dput:
6902d925 2810 dput(dentry);
dae6ad8f
AV		 2811 mutex_unlock(&path.dentry->d_inode->i_mutex);
2812 path_put(&path);
6902d925 2813out_putname:
1da177e4
LT		 2814 putname(from);
2815 return error;
2816}
2817
3480b257 2818SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname)
5590ff0d
UD		 2819{
2820 return sys_symlinkat(oldname, AT_FDCWD, newname);
2821}
2822
1da177e4
LT		 2823int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry)
2824{
2825 struct inode *inode = old_dentry->d_inode;
2826 int error;
2827
2828 if (!inode)
2829 return -ENOENT;
2830
a95164d9 2831 error = may_create(dir, new_dentry);
1da177e4
LT		 2832 if (error)
2833 return error;
2834
2835 if (dir->i_sb != inode->i_sb)
2836 return -EXDEV;
2837
2838 /*
2839 * A link to an append-only or immutable file cannot be created.
2840 */
2841 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
2842 return -EPERM;
acfa4380 2843 if (!dir->i_op->link)
1da177e4 2844 return -EPERM;
7e79eedb 2845 if (S_ISDIR(inode->i_mode))
1da177e4
LT		 2846 return -EPERM;
2847
2848 error = security_inode_link(old_dentry, dir, new_dentry);
2849 if (error)
2850 return error;
2851
7e79eedb 2852 mutex_lock(&inode->i_mutex);
aae8a97d
AK		 2853 /* Make sure we don't allow creating hardlink to an unlinked file */
2854 if (inode->i_nlink == 0)
2855 error = -ENOENT;
2856 else
2857 error = dir->i_op->link(old_dentry, dir, new_dentry);
7e79eedb 2858 mutex_unlock(&inode->i_mutex);
e31e14ec 2859 if (!error)
7e79eedb 2860 fsnotify_link(dir, inode, new_dentry);
1da177e4
LT		 2861 return error;
2862}
2863
2864/*
2865 * Hardlinks are often used in delicate situations. We avoid
2866 * security-related surprises by not following symlinks on the
2867 * newname. --KAB
2868 *
2869 * We don't follow them on the oldname either to be compatible
2870 * with linux 2.0, and to avoid hard-linking to directories
2871 * and other special files. --ADM
2872 */
2e4d0924
HC		 2873SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname,
2874 int, newdfd, const char __user *, newname, int, flags)
1da177e4
LT		 2875{
2876 struct dentry *new_dentry;
dae6ad8f 2877 struct path old_path, new_path;
11a7b371 2878 int how = 0;
1da177e4 2879 int error;
1da177e4 2880
11a7b371 2881 if ((flags & ~(AT_SYMLINK_FOLLOW | AT_EMPTY_PATH)) != 0)
c04030e1 2882 return -EINVAL;
11a7b371
AK		 2883 /*
2884 * To use null names we require CAP_DAC_READ_SEARCH
2885 * This ensures that not everyone will be able to create
2886 * handlink using the passed filedescriptor.
2887 */
2888 if (flags & AT_EMPTY_PATH) {
2889 if (!capable(CAP_DAC_READ_SEARCH))
2890 return -ENOENT;
2891 how = LOOKUP_EMPTY;
2892 }
2893
2894 if (flags & AT_SYMLINK_FOLLOW)
2895 how |= LOOKUP_FOLLOW;
c04030e1 2896
11a7b371 2897 error = user_path_at(olddfd, oldname, how, &old_path);
1da177e4 2898 if (error)
2ad94ae6
AV		 2899 return error;
2900
dae6ad8f 2901 new_dentry = user_path_create(newdfd, newname, &new_path, 0);
1da177e4 2902 error = PTR_ERR(new_dentry);
6902d925 2903 if (IS_ERR(new_dentry))
dae6ad8f
AV		 2904 goto out;
2905
2906 error = -EXDEV;
2907 if (old_path.mnt != new_path.mnt)
2908 goto out_dput;
2909 error = mnt_want_write(new_path.mnt);
75c3f29d
DH		 2910 if (error)
2911 goto out_dput;
dae6ad8f 2912 error = security_path_link(old_path.dentry, &new_path, new_dentry);
be6d3e56
KT		 2913 if (error)
2914 goto out_drop_write;
dae6ad8f 2915 error = vfs_link(old_path.dentry, new_path.dentry->d_inode, new_dentry);
be6d3e56 2916out_drop_write:
dae6ad8f 2917 mnt_drop_write(new_path.mnt);
75c3f29d 2918out_dput:
6902d925 2919 dput(new_dentry);
dae6ad8f
AV		 2920 mutex_unlock(&new_path.dentry->d_inode->i_mutex);
2921 path_put(&new_path);
1da177e4 2922out:
2d8f3038 2923 path_put(&old_path);
1da177e4
LT		 2924
2925 return error;
2926}
2927
3480b257 2928SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname)
5590ff0d 2929{
c04030e1 2930 return sys_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
5590ff0d
UD		 2931}
2932
1da177e4
LT		 2933/*
2934 * The worst of all namespace operations - renaming directory. "Perverted"
2935 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
2936 * Problems:
2937 * a) we can get into loop creation. Check is done in is_subdir().
2938 * b) race potential - two innocent renames can create a loop together.
2939 * That's where 4.4 screws up. Current fix: serialization on
a11f3a05 2940 * sb->s_vfs_rename_mutex. We might be more accurate, but that's another
1da177e4
LT		 2941 * story.
2942 * c) we have to lock _three_ objects - parents and victim (if it exists).
1b1dcc1b 2943 * And that - after we got ->i_mutex on parents (until then we don't know
1da177e4
LT		 2944 * whether the target exists). Solution: try to be smart with locking
2945 * order for inodes. We rely on the fact that tree topology may change
a11f3a05 2946 * only under ->s_vfs_rename_mutex _and_ that parent of the object we
1da177e4
LT		 2947 * move will be locked. Thus we can rank directories by the tree
2948 * (ancestors first) and rank all non-directories after them.
2949 * That works since everybody except rename does "lock parent, lookup,
a11f3a05 2950 * lock child" and rename is under ->s_vfs_rename_mutex.
1da177e4
LT		 2951 * HOWEVER, it relies on the assumption that any object with ->lookup()
2952 * has no more than 1 dentry. If "hybrid" objects will ever appear,
2953 * we'd better make sure that there's no link(2) for them.
e4eaac06 2954 * d) conversion from fhandle to dentry may come in the wrong moment - when
1b1dcc1b 2955 * we are removing the target. Solution: we will have to grab ->i_mutex
1da177e4 2956 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
c41b20e7 2957 * ->i_mutex on parents, which works but leads to some truly excessive
1da177e4
LT		 2958 * locking].
2959 */
75c96f85
AB		 2960static int vfs_rename_dir(struct inode *old_dir, struct dentry *old_dentry,
2961 struct inode *new_dir, struct dentry *new_dentry)
1da177e4
LT		 2962{
2963 int error = 0;
9055cba7 2964 struct inode *target = new_dentry->d_inode;
1da177e4
LT		 2965
2966 /*
2967 * If we are going to change the parent - check write permissions,
2968 * we'll need to flip '..'.
2969 */
2970 if (new_dir != old_dir) {
f419a2e3 2971 error = inode_permission(old_dentry->d_inode, MAY_WRITE);
1da177e4
LT		 2972 if (error)
2973 return error;
2974 }
2975
2976 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
2977 if (error)
2978 return error;
2979
d83c49f3 2980 if (target)
1b1dcc1b 2981 mutex_lock(&target->i_mutex);
9055cba7
SW		 2982
2983 error = -EBUSY;
2984 if (d_mountpoint(old_dentry) || d_mountpoint(new_dentry))
2985 goto out;
2986
3cebde24
SW		 2987 if (target)
2988 shrink_dcache_parent(new_dentry);
9055cba7
SW		 2989 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
2990 if (error)
2991 goto out;
2992
1da177e4 2993 if (target) {
9055cba7
SW		 2994 target->i_flags |= S_DEAD;
2995 dont_mount(new_dentry);
1da177e4 2996 }
9055cba7
SW		 2997out:
2998 if (target)
2999 mutex_unlock(&target->i_mutex);
e31e14ec 3000 if (!error)
349457cc
MF		 3001 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE))
3002 d_move(old_dentry,new_dentry);
1da177e4
LT		 3003 return error;
3004}
3005
75c96f85
AB		 3006static int vfs_rename_other(struct inode *old_dir, struct dentry *old_dentry,
3007 struct inode *new_dir, struct dentry *new_dentry)
1da177e4 3008{
51892bbb 3009 struct inode *target = new_dentry->d_inode;
1da177e4
LT		 3010 int error;
3011
3012 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
3013 if (error)
3014 return error;
3015
3016 dget(new_dentry);
1da177e4 3017 if (target)
1b1dcc1b 3018 mutex_lock(&target->i_mutex);
51892bbb
SW		 3019
3020 error = -EBUSY;
1da177e4 3021 if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry))
51892bbb
SW		 3022 goto out;
3023
3024 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
3025 if (error)
3026 goto out;
3027
3028 if (target)
3029 dont_mount(new_dentry);
3030 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE))
3031 d_move(old_dentry, new_dentry);
3032out:
1da177e4 3033 if (target)
1b1dcc1b 3034 mutex_unlock(&target->i_mutex);
1da177e4
LT		 3035 dput(new_dentry);
3036 return error;
3037}
3038
3039int vfs_rename(struct inode *old_dir, struct dentry *old_dentry,
3040 struct inode *new_dir, struct dentry *new_dentry)
3041{
3042 int error;
3043 int is_dir = S_ISDIR(old_dentry->d_inode->i_mode);
59b0df21 3044 const unsigned char *old_name;
1da177e4
LT		 3045
3046 if (old_dentry->d_inode == new_dentry->d_inode)
3047 return 0;
3048
3049 error = may_delete(old_dir, old_dentry, is_dir);
3050 if (error)
3051 return error;
3052
3053 if (!new_dentry->d_inode)
a95164d9 3054 error = may_create(new_dir, new_dentry);
1da177e4
LT		 3055 else
3056 error = may_delete(new_dir, new_dentry, is_dir);
3057 if (error)
3058 return error;
3059
acfa4380 3060 if (!old_dir->i_op->rename)
1da177e4
LT		 3061 return -EPERM;
3062
0eeca283
RL		 3063 old_name = fsnotify_oldname_init(old_dentry->d_name.name);
3064
1da177e4
LT		 3065 if (is_dir)
3066 error = vfs_rename_dir(old_dir,old_dentry,new_dir,new_dentry);
3067 else
3068 error = vfs_rename_other(old_dir,old_dentry,new_dir,new_dentry);
123df294
AV		 3069 if (!error)
3070 fsnotify_move(old_dir, new_dir, old_name, is_dir,
5a190ae6 3071 new_dentry->d_inode, old_dentry);
0eeca283
RL		 3072 fsnotify_oldname_free(old_name);
3073
1da177e4
LT		 3074 return error;
3075}
3076
2e4d0924
HC		 3077SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname,
3078 int, newdfd, const char __user *, newname)
1da177e4 3079{
2ad94ae6
AV		 3080 struct dentry *old_dir, *new_dir;
3081 struct dentry *old_dentry, *new_dentry;
3082 struct dentry *trap;
1da177e4 3083 struct nameidata oldnd, newnd;
2ad94ae6
AV		 3084 char *from;
3085 char *to;
3086 int error;
1da177e4 3087
2ad94ae6 3088 error = user_path_parent(olddfd, oldname, &oldnd, &from);
1da177e4
LT		 3089 if (error)
3090 goto exit;
3091
2ad94ae6 3092 error = user_path_parent(newdfd, newname, &newnd, &to);
1da177e4
LT		 3093 if (error)
3094 goto exit1;
3095
3096 error = -EXDEV;
4ac91378 3097 if (oldnd.path.mnt != newnd.path.mnt)
1da177e4
LT		 3098 goto exit2;
3099
4ac91378 3100 old_dir = oldnd.path.dentry;
1da177e4
LT		 3101 error = -EBUSY;
3102 if (oldnd.last_type != LAST_NORM)
3103 goto exit2;
3104
4ac91378 3105 new_dir = newnd.path.dentry;
1da177e4
LT		 3106 if (newnd.last_type != LAST_NORM)
3107 goto exit2;
3108
0612d9fb
OH		 3109 oldnd.flags &= ~LOOKUP_PARENT;
3110 newnd.flags &= ~LOOKUP_PARENT;
4e9ed2f8 3111 newnd.flags |= LOOKUP_RENAME_TARGET;
0612d9fb 3112
1da177e4
LT		 3113 trap = lock_rename(new_dir, old_dir);
3114
49705b77 3115 old_dentry = lookup_hash(&oldnd);
1da177e4
LT		 3116 error = PTR_ERR(old_dentry);
3117 if (IS_ERR(old_dentry))
3118 goto exit3;
3119 /* source must exist */
3120 error = -ENOENT;
3121 if (!old_dentry->d_inode)
3122 goto exit4;
3123 /* unless the source is a directory trailing slashes give -ENOTDIR */
3124 if (!S_ISDIR(old_dentry->d_inode->i_mode)) {
3125 error = -ENOTDIR;
3126 if (oldnd.last.name[oldnd.last.len])
3127 goto exit4;
3128 if (newnd.last.name[newnd.last.len])
3129 goto exit4;
3130 }
3131 /* source should not be ancestor of target */
3132 error = -EINVAL;
3133 if (old_dentry == trap)
3134 goto exit4;
49705b77 3135 new_dentry = lookup_hash(&newnd);
1da177e4
LT		 3136 error = PTR_ERR(new_dentry);
3137 if (IS_ERR(new_dentry))
3138 goto exit4;
3139 /* target should not be an ancestor of source */
3140 error = -ENOTEMPTY;
3141 if (new_dentry == trap)
3142 goto exit5;
3143
9079b1eb
DH		 3144 error = mnt_want_write(oldnd.path.mnt);
3145 if (error)
3146 goto exit5;
be6d3e56
KT		 3147 error = security_path_rename(&oldnd.path, old_dentry,
3148 &newnd.path, new_dentry);
3149 if (error)
3150 goto exit6;
1da177e4
LT		 3151 error = vfs_rename(old_dir->d_inode, old_dentry,
3152 new_dir->d_inode, new_dentry);
be6d3e56 3153exit6:
9079b1eb 3154 mnt_drop_write(oldnd.path.mnt);
1da177e4
LT		 3155exit5:
3156 dput(new_dentry);
3157exit4:
3158 dput(old_dentry);
3159exit3:
3160 unlock_rename(new_dir, old_dir);
3161exit2:
1d957f9b 3162 path_put(&newnd.path);
2ad94ae6 3163 putname(to);
1da177e4 3164exit1:
1d957f9b 3165 path_put(&oldnd.path);
1da177e4 3166 putname(from);
2ad94ae6 3167exit:
1da177e4
LT		 3168 return error;
3169}
3170
a26eab24 3171SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname)
5590ff0d
UD		 3172{
3173 return sys_renameat(AT_FDCWD, oldname, AT_FDCWD, newname);
3174}
3175
1da177e4
LT		 3176int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen, const char *link)
3177{
3178 int len;
3179
3180 len = PTR_ERR(link);
3181 if (IS_ERR(link))
3182 goto out;
3183
3184 len = strlen(link);
3185 if (len > (unsigned) buflen)
3186 len = buflen;
3187 if (copy_to_user(buffer, link, len))
3188 len = -EFAULT;
3189out:
3190 return len;
3191}
3192
3193/*
3194 * A helper for ->readlink(). This should be used *ONLY* for symlinks that
3195 * have ->follow_link() touching nd only in nd_set_link(). Using (or not
3196 * using) it for any given inode is up to filesystem.
3197 */
3198int generic_readlink(struct dentry *dentry, char __user *buffer, int buflen)
3199{
3200 struct nameidata nd;
cc314eef 3201 void *cookie;
694a1764 3202 int res;
cc314eef 3203
1da177e4 3204 nd.depth = 0;
cc314eef 3205 cookie = dentry->d_inode->i_op->follow_link(dentry, &nd);
694a1764
MS		 3206 if (IS_ERR(cookie))
3207 return PTR_ERR(cookie);
3208
3209 res = vfs_readlink(dentry, buffer, buflen, nd_get_link(&nd));
3210 if (dentry->d_inode->i_op->put_link)
3211 dentry->d_inode->i_op->put_link(dentry, &nd, cookie);
3212 return res;
1da177e4
LT		 3213}
3214
3215int vfs_follow_link(struct nameidata *nd, const char *link)
3216{
3217 return __vfs_follow_link(nd, link);
3218}
3219
3220/* get the link contents into pagecache */
3221static char *page_getlink(struct dentry * dentry, struct page **ppage)
3222{
ebd09abb
DG		 3223 char *kaddr;
3224 struct page *page;
1da177e4 3225 struct address_space *mapping = dentry->d_inode->i_mapping;
090d2b18 3226 page = read_mapping_page(mapping, 0, NULL);
1da177e4 3227 if (IS_ERR(page))
6fe6900e 3228 return (char*)page;
1da177e4 3229 *ppage = page;
ebd09abb
DG		 3230 kaddr = kmap(page);
3231 nd_terminate_link(kaddr, dentry->d_inode->i_size, PAGE_SIZE - 1);
3232 return kaddr;
1da177e4
LT		 3233}
3234
3235int page_readlink(struct dentry *dentry, char __user *buffer, int buflen)
3236{
3237 struct page *page = NULL;
3238 char *s = page_getlink(dentry, &page);
3239 int res = vfs_readlink(dentry,buffer,buflen,s);
3240 if (page) {
3241 kunmap(page);
3242 page_cache_release(page);
3243 }
3244 return res;
3245}
3246
cc314eef 3247void *page_follow_link_light(struct dentry *dentry, struct nameidata *nd)
1da177e4 3248{
cc314eef 3249 struct page *page = NULL;
1da177e4 3250 nd_set_link(nd, page_getlink(dentry, &page));
cc314eef 3251 return page;
1da177e4
LT		 3252}
3253
cc314eef 3254void page_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie)
1da177e4 3255{
cc314eef
LT		 3256 struct page *page = cookie;
3257
3258 if (page) {
1da177e4
LT		 3259 kunmap(page);
3260 page_cache_release(page);
1da177e4
LT		 3261 }
3262}
3263
54566b2c
NP		 3264/*
3265 * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS
3266 */
3267int __page_symlink(struct inode *inode, const char *symname, int len, int nofs)
1da177e4
LT		 3268{
3269 struct address_space *mapping = inode->i_mapping;
0adb25d2 3270 struct page *page;
afddba49 3271 void *fsdata;
beb497ab 3272 int err;
1da177e4 3273 char *kaddr;
54566b2c
NP		 3274 unsigned int flags = AOP_FLAG_UNINTERRUPTIBLE;
3275 if (nofs)
3276 flags |= AOP_FLAG_NOFS;
1da177e4 3277
7e53cac4 3278retry:
afddba49 3279 err = pagecache_write_begin(NULL, mapping, 0, len-1,
54566b2c 3280 flags, &page, &fsdata);
1da177e4 3281 if (err)
afddba49
NP		 3282 goto fail;
3283
1da177e4
LT		 3284 kaddr = kmap_atomic(page, KM_USER0);
3285 memcpy(kaddr, symname, len-1);
3286 kunmap_atomic(kaddr, KM_USER0);
afddba49
NP		 3287
3288 err = pagecache_write_end(NULL, mapping, 0, len-1, len-1,
3289 page, fsdata);
1da177e4
LT		 3290 if (err < 0)
3291 goto fail;
afddba49
NP		 3292 if (err < len-1)
3293 goto retry;
3294
1da177e4
LT		 3295 mark_inode_dirty(inode);
3296 return 0;
1da177e4
LT		 3297fail:
3298 return err;
3299}
3300
0adb25d2
KK		 3301int page_symlink(struct inode *inode, const char *symname, int len)
3302{
3303 return __page_symlink(inode, symname, len,
54566b2c 3304 !(mapping_gfp_mask(inode->i_mapping) & __GFP_FS));
0adb25d2
KK		 3305}
3306
92e1d5be 3307const struct inode_operations page_symlink_inode_operations = {
1da177e4
LT		 3308 .readlink = generic_readlink,
3309 .follow_link = page_follow_link_light,
3310 .put_link = page_put_link,
3311};
3312
2d8f3038 3313EXPORT_SYMBOL(user_path_at);
cc53ce53 3314EXPORT_SYMBOL(follow_down_one);
1da177e4
LT		 3315EXPORT_SYMBOL(follow_down);
3316EXPORT_SYMBOL(follow_up);
3317EXPORT_SYMBOL(get_write_access); /* binfmt_aout */
3318EXPORT_SYMBOL(getname);
3319EXPORT_SYMBOL(lock_rename);
1da177e4
LT		 3320EXPORT_SYMBOL(lookup_one_len);
3321EXPORT_SYMBOL(page_follow_link_light);
3322EXPORT_SYMBOL(page_put_link);
3323EXPORT_SYMBOL(page_readlink);
0adb25d2 3324EXPORT_SYMBOL(__page_symlink);
1da177e4
LT		 3325EXPORT_SYMBOL(page_symlink);
3326EXPORT_SYMBOL(page_symlink_inode_operations);
d1811465 3327EXPORT_SYMBOL(kern_path);
16f18200 3328EXPORT_SYMBOL(vfs_path_lookup);
f419a2e3 3329EXPORT_SYMBOL(inode_permission);
1da177e4
LT		 3330EXPORT_SYMBOL(unlock_rename);
3331EXPORT_SYMBOL(vfs_create);
3332EXPORT_SYMBOL(vfs_follow_link);
3333EXPORT_SYMBOL(vfs_link);
3334EXPORT_SYMBOL(vfs_mkdir);
3335EXPORT_SYMBOL(vfs_mknod);
3336EXPORT_SYMBOL(generic_permission);
3337EXPORT_SYMBOL(vfs_readlink);
3338EXPORT_SYMBOL(vfs_rename);
3339EXPORT_SYMBOL(vfs_rmdir);
3340EXPORT_SYMBOL(vfs_symlink);
3341EXPORT_SYMBOL(vfs_unlink);
3342EXPORT_SYMBOL(dentry_unhash);
3343EXPORT_SYMBOL(generic_readlink);
kernel source for telekom puls (alcatel/mediatek)